Skip to content

Wiki log

Chronological record of wiki maintenance and notable ingest events (newest first).

2026-06-06

2026-06-04

  • projects/education-startup-ux — day job, not personal portfolio: Reclassified employer parent-app work — removed from personal Exploratory & background roster on index; new Day job (context only) section. Updated projects/education-startup-ux (Type + Day job vs personal portfolio section), What-Im-Working-On, goals (alignment table + Seat 4 + distillation skill note), Grok-Brief (ultra-minimal + routing + education capsule), glossary (Boxy), mkdocs.yml nav label. Reason: education startup is day job, not a personal side project (e.g. not a personal Gemini XPRIZE / hackathon lane).

2026-06-02

  • projects/happy-body — mentor profile (Jerzy website): Added Mentor profile (Jerzy website) section — practical/direct bio for Jerzy’s mentor roster (Tim Ferriss + Naval discovery, PNW active lifestyle, micro-progression, injury wisdom, food timing, daily accountability). Cross-updated Key features mentoring bullet; this log.

2026-05-30

  • projects/movie-taste-app — new project page: Ingested Movie Taste Tracker (~/Projects/movie-taste-app, GitHub dmoniker/movies): dual-profile TMDB watch journal, taste radar, Grok recommendation re-ranking, JSON backup, and Netflix viewing-history CSV import (TMDB match → mark seen). Cross-updated: index, tools-and-repos, mkdocs.yml, this log.

2026-05-21

2026-05-20

  • projects/crusty — Hermes cutover + Aye Robot re-validation: Documented OpenClaw → Hermes cutover complete on the dedicated laptop (2026-05-20); SuperGrok OAuth active. In flight: experiment until morning Aye Robot works reliably under Hermes (Telegram + xurl + grok-imagine). Cross-updated: projects/crusty, What-Im-Working-On (snapshot 2026-05-20, four parallel priorities), index, this log.

  • projects/lunarcast — subscription persistence maintenance: Logged active fix — app reloads subscribed shows too often; target local cache first, background RSS refresh, library usable offline after one sync. Growth remains paused. Cross-updated: projects/lunarcast, What-Im-Working-On, index.

2026-05-18

  • grok-agent-architecture-may-2026 — Grok + agent architecture (Skills, Build, multi-agent, tiers, connectors): New canonical wiki page (May 2026 posture) tying Hermes + SuperGrok OAuth on Crusty to cloud Grok (Skills/computer-use/multi-agent synthesis) plus planned WikiMaintainer loop (Hermes-scoped Git → human approval → push → Vercel → wiki.darcymenard.com). Cross-updated: projects/crusty (hybrid bullet + Related), projects/personal-projects-wiki (deploy URL, upkeep plan, MkDocs/Vercel stack, Related tidy), Grok-Brief (Grok ecosystem snapshot + phrase row + When I don’t paste hint + maintainer footer), What-Im-Working-On (wiki hosting line + snapshot Crusty bullet + tooling row), index (Last updated + Supporting pages row), tools-and-repos (Projects Wiki row), SCHEMA.md (canonical file bullet + hosting tip), mkdocs.yml (nav).

2026-05-17

  • Main Mac — local LLM (public-safe summary): Documented LM Studio on the main personal Mac (M2 Pro, 16 GB, 2023 MacBook Pro class) running Google Gemma 4 instruction-tuned MLX weights (4-bit quant) for sensitive / local-only text kept off typical cloud chat/API paths — orthogonal to Crusty (Hermes / Grok); explicit nuance that local inference ≠ full privacy (telemetry, paste discipline). Cross-updated: What-Im-Working-On (Tools & stack), personal-operating-system (Portable layers + meta Last updated), projects/crusty (Note to LLMs), index (Last updated + retrieval note), SCHEMA.md (maintainer LLM note), this log.

2026-05-16

  • Three parallel priorities + Crusty → Hermes: What-Im-Working-On snapshot 2026-05-16 — active lines: (1) Happy Body coaching prep — Jerzy mentoring course, iOS content (exercise + flower + Thais), mentoring path; (2) Rusty Cage / site — three finished recordings (Caterpillar Dream, Tangerine Sky, Technology) + terracain.com, gig FB/list capture still open; (3) Crusty — Hermes Agent replacing OpenClaw, SuperGrok via xai-oauth (xAI, Hermes OAuth), hermes claw migrate, re-validate Telegram + xurl. projects/crusty title/type/description + Grok personal section (Hermes OAuth paragraph, irreplaceable bullets); Grok-Brief ultra-minimal + Crusty capsule + phrase row; index Last updated + retrieval note + crusty row; projects/openclaw-autonomy-org deployment + tech + harness; What-Im-Working-On tools line + “How to use” step 3; SCHEMA.md (maintainer LLM / Crusty handoff).

2026-05-14

  • projects/rusty-cage — originals + terracain + gig capture: Immediate next = three finished Rusty Cage recordings — Caterpillar Dream, Tangerine Sky, Technology — with https://terracain.com cross-promotion on each release. Reminder logged: better mechanism for Facebook (or mailing-list) signups when playing gigs — pick one default path (QR, short URL, stage line), test at next show. Gig table: 2026-05-01 Anderson played; 2026-07-25 Third Place Commons still upcoming. Cross-updated What-Im-Working-On (snapshot 2026-05-14, Rusty bullet, pain point), Grok-Brief (ultra-minimal + Rusty Cage + Happy Body capsules), tools-and-repos (Terracain row), index Last updated + life table.

  • projects/happy-body — Jerzy course + iOS content bundle: Parallel priority = complete Jerzy’s mentoring course (supplied materials). Native iOS app: add exercise videos, flower video for relaxation, and Thais (Meditation from Thaïs) for cooldown / meditation playback (alongside existing bounded assessment-upload posture). Cross-updated What-Im-Working-On, Grok-Brief, index, this log.

2026-05-10

  • projects/crusty — Crusty surface narrowed; mail + meals on grok.com: Unattended OpenClaw dailies on Crusty = morning Aye Robot only. Newsletter / rustycagetrio mail summary and 3 pm meal ideas migrated to grok.com scheduled tasks on Darcy’s Grok account (Gmail connector). Prior Planned shift (grok.com on the Crusty laptop for band mail) superseded. Cross-updated: projects/crusty, Grok-Brief, What-Im-Working-On, goals, personal-operating-system, projects/openclaw-autonomy-org, projects/rusty-cage, tools-and-repos, projects/operator-agent, index Last updated.

  • projects/ayerobot-comic — Crusty vs Grok-task mini experiment: Documented parallel run (Crusty scheduled Aye Robot pipeline vs grok.com scheduled Aye Robot task) to compare creative quality and reliability; Waypoint 1 remains concluded — informal bake-off only; X posting still xurl when automated. Cross-updated Grok-Brief (ultra-minimal + Aye Robot + Aye Robot comic capsules), What-Im-Working-On (snapshot date + bullet), projects/crusty (daily table), index Last updated + ayerobot row.

  • projects/crusty — grok.com on Crusty laptop (planned) — superseded same day (late): Original note envisioned grok.com on the dedicated machine for Rusty Cage mail. Later 2026-05-10: band summary + meals instead run on grok.com under Darcy’s Grok account; Crusty dropped those jobs — see newer log entry above. (Kept for chronology.)

2026-05-05

  • projects/music-production — diatonic keys + Low F: Set Low F (not mid/high F), G, A, B♭, C; project page notes longer reeds / heavier breath on the low harp. Cross-updated Grok-Brief, What-Im-Working-On, index Last updated + row.

  • projects/music-production — harmonica technique path: Documented progression (1st position: clean singles, blow/draw rhythms) → draw bends → 2nd / cross harp + 3rd / slant harp; current = muscle memory + musicality in cross (major + minor) and slant; 12th position + overblows deferred. Cross-updated What-Im-Working-On (music practice bullet), Grok-Brief (ultra-minimal music sentence + Music production capsule), index music-production row.

  • Music / wellness cross-links: Expanded projects/music-production (harmonica as breath + patience + lung capacity practice; stress relief; blues singing therapeutic; optional how-to videos, saturated market). projects/happy-body now notes music as adjacent happy-body rhythm. projects/rusty-cage goals → ~two gigs/month + rehearsal every other week; Related → happy-body. What-Im-Working-On snapshot 2026-05-05; Grok-Brief ultra-minimal + Rusty Cage + Music production capsules + phrase row (harmonica / blues harp). index life-table rows updated.

  • Grok personal (4.3) — Crusty use-case overlap documented (revised same session): Personal Grok subscription now running Grok 4.3 with Gmail, Calendar, GitHub, custom MCP connectors via URL, up to 10 independent scheduled tasks, and the ability to pass the wiki URL as context to any task. Initial read identified autonomous scheduling as the primary gap vs Crusty; that gap is closed for simple routines. Revised net verdict: Grok personal now covers the simple daily routine layer (~60–70% of Crusty's use case volume by task count) — scheduled mail digest, meal ideas, calendar summaries, GitHub roll-ups — with full project context via wiki URL. Irreplaceable Crusty surface narrows to three things: (1) posture/identity separation — Operator Agent ops and any real-P&L work cannot run through personal accounts; (2) Telegram approval gate — the draft → human-gate → execute → verify pattern requires an agent that can push draft actions unprompted and wait for a Telegram reply before executing; (3) X posting via xurl — Grok personal has no X API posting surface. Also documented: personal subscription running Grok 4.3 is not the metric for the Crusty orchestrator upgrade — the gate is the 4.3 API rate at console.x.ai. Open strategic question logged: whether Crusty's simple routines should migrate to Grok personal, letting Crusty specialize on posture-sensitive and gate-pattern use cases; revisit at 2026-07-18. Cross-updated: projects/crusty (new section, revised), What-Im-Working-On (Crusty line), Grok-Brief (Crusty capsule).

2026-05-03

  • Both waypoints concluded; EV L2 POC killed (2026-05-03): Waypoint 1 (projects/ayerobot-comic) — experiment stopped intentionally after proving gate pattern end-to-end (brainstorm → Telegram approve → xurl post). Waypoint 2 / EV L2 Shoreline POC (projects/ev-charging-shoreline-poc) — killed after full economic stress-test revealed median net returns (~$50–$100/mo per charger) insufficient for capital, 80–150 hrs Year 1 owner time, and illiquidity vs preserving existing equity positions. Decision: do not sell stocks to fund this; capital stays invested. Operator Agent thesis and long-term ladder (laundromat → RaaS → own hardware) remain intact; next node TBD at 2026-07-18 review. Focus window continues with direction open. Cross-updated: projects/ayerobot-comic (status → concluded), projects/operator-agent (Node 1 killed, waypoint 2 closed), projects/ev-charging-shoreline-poc (status → archived), What-Im-Working-On (waypoints section rewritten), index (top-priority table + archived table + last-updated line).

  • Operator Agent — EV L2 Node 1 (Shoreline POC): New projects/ev-charging-shoreline-poc (multifamily-first Level 2 economics, Seattle City Light / grant stacking, PlugShare scouting, Crusty-friendly ops cadence, 10-year demand note). Parent projects/operator-agent reworked: Node 1 = EV POC, smart vending sprint archived (weekly restock incompatible with automation posture); long-term ladders scaled L2 → laundromat → RaaS → own hardware; Crusty Day 1 skills reframed (charger telemetry, settlement mismatches, landlord PM drafts — SKU PO parked). Cross-updated What-Im-Working-On, projects/crusty, goals (Seat 2 wording + Operator alignment row + curriculum item 5), Grok-Brief (Waypoint 2, routing table, Operator + Crusty Direction capsules), index, MkDocs nav (ev-charging-shoreline-poc), projects/buddyline framing.

2026-05-01

  • projects/personal-health-diet — DTC lab workflow documented: Under Direct-pay labs, added Workflow — what you’re actually buying (default you order → PSC → PDF; lab QA vs clinical interpretation; generic requisition physician vs real review; optional vendor telehealth; intentional handoff to treating clinician). Practical tips lead with self-serve framing.

  • projects/personal-health-diet — baseline lab gaps + direct-pay channels: New sections Baseline labs — gaps vs typical annual wellness (already‑covered annual list; prioritized Lp(a), apoB, fasting insulin, hs‑CRP, UACR, ferritin/iron, 25‑OH D timing, testosterone/SHBG; optional second tier; carotid ultrasound imaging note) and Direct-pay labs — reputable, usually faster than yearly PCP turnaround (Quest Health, Labcorp OnDemand, Walk-In Lab, Life Extension; tips on à la carte vs bundles, HSA/FSA, trending). Not medical advice.

  • projects/supplements-stack — Nusava label on file: Photo wiki/assets/nusava-organic-d3-k2-liquid.png; Supplement Facts synced in wiki (10k IU D₃; K₂ 60 mcg MK‑4 + 60 mcg MK‑7; 250 mg ALA; organic chia / MCT / olive oils); label shows USDA Organic, Non‑GMO, GMP, Oregon Tilth. Grok-Brief + projects/personal-health-diet text aligned (was briefly MK‑7‑only/MCT‑only shorthand).

  • projects/supplements-stack / projects/personal-health-diet — liquid D₃ + K₂ (+ MCT): Replaced Kirkland vitamin D₃ pills with 1 ml (full dropper) liquid D₃ + K₂ in morning coffee (10,000 IU D₃ + 120 mcg K₂ with oil carriers) with 10 g creatine; rest of morning capsules unchanged. Rationale on projects/supplements-stack and Tier 1 on projects/personal-health-diet; 25‑OH vitamin D follow-up ~6–8 weeks and ~40–60 ng/mL target documented. Grok-Brief ultra‑minimal + Supplement stack capsule aligned.

  • projects/personal-health-diet — May 2026 health optimization backlog: Added Health optimization — what’s next — Tier 0 (sleep / training+creatine+protein / TRE already running), sequenced Tier 1–4 (K₂+D₃, labs + imaging triggers, nattokinase/glycine/berberine/red-light/sauna, NMN–OneSkin tier, aspirin/metformin/peptide deferrals), 90-day rollout, operating rules (≤2–3 new items, labs first, clinician for higher-risk agents). projects/supplements-stack gained Queued additions table (K₂ immediate; Tier 2 nattokinase/glycine/berberine planned).

  • Grok-Brief — supplement stack cold-start: Ultra-minimal paragraph + Phrase → project row + Supplement stack capsule (dose table, routing aliases) + When I don’t paste line; footer maintainer list now includes Supplement stack updates.

  • New page — projects/supplements-stack: Documented morning supplement stack (products + doses: 2× vitamin D3 2000 IU, 1× each of niacinamide, omega-3, organic multivitamin, Qunol magnesium L-threonate/citicoline, Nature Made magnesium). Photo stored at wiki/assets/morning-supplement-stack.png. Cross-links from projects/personal-health-diet; index life table row; MkDocs nav under Projects. User treats this as non-sensitive for public wiki. Evidence subsection stub for future study notes.

  • projects/personal-context-ios — product pitch + names (Context / LocalMind): Elevator pitch on project page — local Markdown folder, retrieve only relevant notes, Apple on‑device default, preview context before cloud, one‑tap Grok / Claude / BYOK, optional iCloud wiki sync; privacy‑first story tied to Apple's on‑device roadmap. Naming collision / App Store search risk logged on project page (Context generic; LocalMind more distinctive). Cross-updated Grok-Brief (routing + capsule), index row, What-Im-Working-On one‑liner; index Last updated bumped to align with SCHEMA “index-row change” signal (historical wiki snapshot headings on other pages unchanged). Real‑world ingest date label 2026-01-31, merged here under ## 2026-05-01 to preserve newest‑first dated sections discipline. No priority change — still exploratory and outside the 90‑day focus window.

  • projects/personal-context-ios business model — local-first MVP before API resale: Added a Business model / API cost section. API-cost passthrough is now recorded as an unresolved product-shape risk: reselling inference would add metering, abuse prevention, margin risk, usage-surprise support, and App Store subscription complexity. Default v1: do not become an inference reseller. Recommended MVP order is now (1) local-first / Apple Intelligence-first for capture cleanup, classification, summarization, redaction, sensitivity labeling, and draft file edits; (2) BYOK / power-user mode with provider keys stored in Keychain; (3) provider-subscription passthrough / prompt packaging; (4) limited included credits only later after usage data. No priority change — still exploratory and outside the 90-day focus window.

  • projects/personal-context-ios refined — private context firewall + broader use cases: Added a Trust posture — private context firewall section. The product wedge is not "send your deepest medical / relationship history to model APIs"; raw personal context should stay local, sensitivity should be classified, and only approved summaries / selected snippets should leave the phone. Captured the six-step posture: local markdown, sensitivity classification, minimum-context retrieval, show what will be sent, prefer on-device summarization/redaction, send only approved context externally. Added general use cases beyond project wikis: preference memory, decision journal / values layer, life admin memory, coaching memory, conversation prep, portable prompt profile, and voice-capture inbox that proposes file edits. No priority change — still exploratory and outside the 90-day focus window.

  • Strategic refinement — apps as trusted clients, not generic UI plays: Updated goals to clarify that the north-star thesis is not "apps are dead." Standalone apps can still be viable income paths when the phone is the permissioned client for a trusted workflow (camera, microphone, HealthKit, Photos, local storage, push, playback, offline use, family/device permissions, on-device processing). The weak bet remains pure UI polish; the strong bet is app-as-trust-surface for a context-rich workflow or agent. Added projects/happy-body to the goals alignment table as the clearest current example (native iOS video assessment / coaching loop). Updated projects/happy-body with the app-as-trust-surface rationale, and projects/personal-context-ios with the Apple Intelligence / OS-memory platform risk plus the narrower defensible wedge: owned markdown, BYOK / multi-provider choice, explicit retrieval controls, and approved edits back to files. No project priorities changed.

  • New exploratory project — projects/personal-context-ios: Native iOS idea for a personal context / memory client (markdown vault: soul.md, memory.md, …), retrieval into any LLM, BYOK (no author inference cost); Apple on-device as privacy-first option. Friction: Mac Cursor maintains this wiki; Grok on phone/car has no repo — app would close the loop (talk → approve → canonical files). Prototype path: Cursor + wiki folder + rules before a dedicated UI. Auth: OpenAI no clean OAuth for third-party clients; unofficial session-token flows gray area — wiki records prefer BYOK. Cross-links: projects/personal-projects-wiki, projects/llm-maintained-context, Grok-Brief, What-Im-Working-On, index.

2026-04-30

  • Crusty — model posture clarified; Happy Body — native iOS video: projects/crusty Steady-state economics now states explicitly that grok-4-1-fast-reasoning is the intentional affordable steady state (not ambiguity vs the historical two-tier doc); premium orchestrator stays parked until Grok 4.3 (or xAI’s then-current fuller tier) is cheap enough. projects/happy-body, Grok-Brief, What-Im-Working-On, index, projects/operator-agent updated for Happy Body: native iOS app with video as the product rationale for a standalone app; limited uploads acceptable for flexibility assessments; web at happybodygrok.com for brand/funnel.

2026-04-29

  • Crusty — economics rollback + projects/carlot-chat archived: projects/crusty gained Steady-state economics (2026-04-29) — grok-4-1-fast-reasoning again the primary workload tier; grok-4.20-0309-reasoning orchestrator routing paused because Aye Robot generation volume burned too many tokens. Documented tradeoff (premium orchestrator hedge vs spend), intermittent morning Telegram image, subjective idea-quality drift, cloud vs local revisit held until June 2026 (M5 / Apple AI). Cross-updated Grok-Brief, What-Im-Working-On, index, recurring-issues, projects/openclaw-autonomy-org, projects/aye-robot-crusty-paused-x-automation, projects/ayerobot-comic, projects/buddyline. projects/carlot-chat → Archived — client won't invest for now; similar SMB client-chat work explicitly deprioritized vs Waypoints 1 + 2. goals v1.5: alignment row + Seat 2b column + Two candidate income seats paragraph revised.

  • projects/crusty Model routing H2 — execution complete: Renamed the section from execution underway to execution complete (decision locked 2026-04-27 — still via Claude Code on Crusty). Fragment: #model-routing-decision-locked-2026-04-27--execution-complete-via-claude-code-on-crusty. Updated in-page #... links and personal-operating-system; historical log entry (morning cron bullet) Anchors fixed line edited to match the new title so the chronicle does not point at a non-existent heading slug.

  • Wiki alignment — two former blockers explicit site-wide: The wiki previously treated (1) Crusty orchestrator model routing (premium grok-4.20-0309-reasoning to avoid confabulation on gate judgment) and (2) xurl on Crusty (install + OAuth 2.0 PKCE with the dev app — official path replacing post_comic_test.py for native media) as the gating work before a trustworthy live comic cadence. Both were already completed 2026-04-27; this pass updates projects/operator-agent, projects/buddyline, projects/aye-robot-crusty-paused-x-automation, index, Grok-Brief, projects/crusty, and recurring-issues so no page still lists them as Remaining / top-priority todo. Next Waypoint 1 work remains mentions listen path + post-path reliability (~1 week).

  • Aye Robot — next concrete step named (mentions listen path): The wiki now treats projects/ayerobot-comic#next-concrete-step--mentions-listen-path as the canonical spec for the listen build: scheduled xurl/X API read of @ayerobotcomic mentions (hourly cadence in the cost model) → Telegram digest to Darcy → optional Telegram-gated xurl reply; out of scope — browser automation, auto-like/follow/quote. In parallel (~1 week), harden the post path for daily reliability (same brainstorm → approve → xurl loop already proven). Updated: projects/ayerobot-comic (new section + status), projects/aye-robot-crusty-paused-x-automation (status + handoff step 6), What-Im-Working-On, index, Grok-Brief (ultra-minimal, Waypoint 1, Aye Robot + Aye Robot comic capsules, Crusty daily jobs).

  • Aye Robot — gate unlocked (E2E live): Darcy ran the full Crusty pipeline successfully: brainstorm → select best idea → Telegram → manual approve → xurl post to @ayerobotcomic on X. The composite Aye Robot gate (Crusty two-tier model routing Phases 3–4 + Telegram human-gate + first live post on the xurl path) is no longer blocking Waypoint 1. Wiki updated: What-Im-Working-On, index, Grok-Brief, projects/ayerobot-comic, projects/crusty, projects/aye-robot-crusty-paused-x-automation, projects/openclaw-autonomy-org; recurring-issues — Crusty single-model stack moved to Resolved; open-loop bullets reframed to mentions + daily cadence. Prior same-day log entry (morning cron + routing research) remains below.

  • Crusty / Aye Robot — morning image cron + wiki alignment with routing research: Darcy updated Crusty's cron so a fresh Aye Robot panel is generated and delivered on Telegram every morning (2026-04-27). The remaining Waypoint 1 automation build is not the image clock — it is the auto-generated X-post package (caption + image in the same Telegram thread) so Darcy can manually approve before Crusty runs xurl to post. Sequencing on projects/crusty: model routing is decision-locked two-tier; Phase 1–3 (Claude Code on Crusty) in flight, Phase 4 dry-run, then xurl path. Added projects/crusty Research notes (2026-04-27): native OpenClaw per-agent + SKILL.md routing, rejection of multi-provider LLM proxy layers (e.g. Terraphim-style) for xAI-only Crusty, Alex Finn chief-of-staff prior-art cross-check, pointer that the verbatim long Claude Code paste-block stays session-side / the numbered Sequencing list is the contract. Anchors fixed: in-repo links to the old Model routing (open decision — top priority) slug now point at Model routing (decision locked 2026-04-27 — execution complete); personal-operating-system chief-of-staff cross-ref updated. Aye Robot pages (projects/ayerobot-comic, projects/aye-robot-crusty-paused-x-automation): status + "still open" blocks now distinguish image scheduling LIVE vs X-post draft + approval vs xurl; handoff step 0 no longer says "split into two OpenClaw sessions" (obsolete). projects/openclaw-autonomy-org Hermes section: gating-question #1 resolved yes (native routing); Decision bullets 1–3 and revisit trigger (b) updated so Hermes is deferred for complexity, not an unresolved OpenClaw capability; Aye Robot feature line reflects cron + next build. index + What-Im-Working-On snapshot 2026-04-27: Waypoint 1 row, pain points, wins, and Crusty blurb. Grok-Brief updated (ultra-minimal + Crusty capsule lines) for 2026-04-27 Aye Robot + routing state — still no operational file paths. Not in wiki: full verbatim multi-KB Claude Code paste-block (drift risk; projects/crusty Sequencing + Research notes are the stable surface).

2026-04-26

  • Crusty — Claude Code installed (debug-time force multiplier landed on the Crusty machine): Closed the Claude Code on Crusty TODO that had been the lead pain-point on What-Im-Working-On since 2026-04-21. Install was performed directly on Crusty in one session per the public setup checklist on projects/crusty: fresh Anthropic Console account on Crusty's band identity (email + password + email-link verification + TOTP 2FA — SSO declined to keep each Crusty service independently auth'd; trade-off analysis for SSO captured in this chat but kept out of the wiki since the decision is now made), payment method added, $10/mo hard spend cap in place, API key created, native installer (curl -fsSL https://claude.ai/install.sh | bash per docs.claude.com) landed Claude Code 2.1.119 at ~/.local/bin/claude (symlink into ~/.local/share/claude/versions/<version>/), key paste through interactive first-run prompt to Keychain (no ANTHROPIC_API_KEY on disk in any shell rc — verified with grep across ~/.zshrc ~/.zprofile ~/.bashrc ~/.bash_profile ~/.profile), default model pinned per recommendation. One deviation from the original 2026-04-21 setup checklist worth logging because it's a permanent posture concession: existing credit card used on Console billing instead of the recommended prepaid Visa. Card-graph collision at Anthropic's billing side is now permanent; defended by the $10 spend cap as the hard bill-shock ceiling. Three resolved open handoff questions (originally listed on projects/crusty as Q1–Q3 of the Open handoff questions block): (Q1) canonical macOS install path is curl | bash per current docs.claude.com/en/docs/claude-code/setup, not the older npm i -g @anthropic-ai/claude-code the original checklist mentioned — the official docs have drifted as the original framing predicted; (Q2) no .env / shell-rc ingestion observed on first run — credential lands in Keychain only, hygiene grep returned clean, the cwd-scoping discipline still applies for actual debug sessions but is about narrowing context surface, not credential leak; (Q3) Haiku vs Sonnet — pinned Haiku as the default per the original suggestion, with Sonnet via /model for harder OpenClaw config work. Q4 (30-day spend revisit) now scheduled 2026-05-26 — if observed monthly spend trends over ~$15/mo, the API-key-vs-Pro cost calculus flips and Pro becomes the better dollar deal; under that, Pro's day-one mitigations (band identity, claude.ai history off, TOTP 2FA, two-step revocation) listed on projects/crusty get applied. Two install-time gotchas captured in the as-built section of projects/crusty for future fresh-machine / disaster-recovery setups: (a) the native installer does not always add ~/.local/bin to PATH despite claiming to — fix is echo 'export PATH="$HOME/.local/bin:$PATH"' >> ~/.zshrc and open a new terminal window, not exec zsh mid-paste (exec zsh discards any queued input that follows it); (b) checklist now flags this PATH gotcha explicitly so future-Darcy isn't debugging a "command not found" against a binary that actually installed correctly. Invariant preserved: "Claude runs on Crusty, but Crusty doesn't run on Claude" — Claude Code is dev-time only, scoped to Darcy's interactive shell, never wired into OpenClaw skills; runtime autonomous loop stays 100% Grok per the original 2026-04-21 commitment. Two posture choices preserved that aren't deviations but are worth re-naming because they were the right call: (1) Console-only path means no claude.ai chat-surface identity attached to Crusty's graph; (2) email+password+TOTP rather than SSO keeps the band-identity SSO host's blast radius from extending to Anthropic billing in the event of a band-identity SSO compromise — at the cost of one more credential to manage, which the band-identity password manager on Crusty's local user account holds. Updated: projects/crusty (Claude Code section heading flipped from TODO — force multiplier to installed 2026-04-26 — debug-time force multiplier, status block rewritten as as-built, Recommendation header updated to validated 2026-04-26 — followed with one deviation, Setup checklist reframed as executed 2026-04-26 — preserved here as disaster-recovery template with as-built notes block at the top covering the credit-card deviation + install-command update + PATH gotcha + hygiene-check pass, Open handoff questions Q1–Q3 marked Resolved with answers inline + Q4 30-day spend revisit scheduled, Sequencing section updated to "(1) install Claude Code — done 2026-04-26; (2) use it to work through the model-routing config edits — now active; (3) resume Aye Robot Waypoint 1 — gated on (2)", Known issues bullet flipped from TODO to installed-with-cross-ref-to-section); index (last-updated bumped to 2026-04-26; Open TODO line under Top-priority replaced with Force-multiplier installed line keeping the cost cap + invariant + 30-day-revisit hooks; Crusty status row in the top-priority sub-table updated to "model-routing decision open 2026-04-21; Claude Code installed 2026-04-26"); What-Im-Working-On (snapshot date bumped to 2026-04-26; Pain points / open loops top bullet removed since the TODO is closed; Wins / momentum gained a Claude Code bullet covering install state + invariant + 30-day revisit); projects/openclaw-autonomy-org (the Hermes-deferral Why this is the right call now line that listed three unblocked critical-path items is now two — Claude Code TODO closed, Model routing + xurl install + live tweet still active — with a parenthetical preserving the closed item for change-history readability). Not updated: Grok-Brief (the Claude Code install is a Crusty-machine posture state, not cold-start context Grok needs in a blank thread — same reason the deferred BuddyLine and creative-space pages aren't in Grok-Brief either); recurring-issues (the openclaw.zsh compdef ordering warning that surfaced during install — ~/.openclaw/completions/openclaw.zsh:3688: command not found: compdef printed every shell launch on Crusty — is real but cosmetic, fixable by moving the OpenClaw source line after autoload -Uz compinit && compinit in Crusty's shell rc; that's Crusty-machine operational specifics so it lives in private notes per the public-wiki discipline, not on recurring-issues); goals (no fleet-roster change — Claude Code on Crusty is debug-time tooling, not a new agent seat); personal-operating-system (the 7-layer cross-walk already names Claude Code as a possible skills / automations layer harness alongside Cursor and OpenClaw — the install fits the existing right-hand column, no vocabulary change needed). No active-project status changes beyond the Claude Code TODO closure itself: Waypoint 1 still gated on the Crusty model-routing decision, Waypoint 2 sprint still on its 2026-05-02 go/no-go, Carlot conditional unchanged. Sequencing payoff confirmed: the Claude-Code-first-then-routing order argued for in the original projects/crusty Sequencing relative to the model-routing decision section now plays out — routing-decision execution can use Claude Code on Crusty in-place rather than handoff-text iteration from main Mac, which is the entire force-multiplier thesis the install was scoped against.

2026-04-25

  • index gained a manual "Last updated" line (option A of three considered): Added a single italic line under the index intro — "Last updated: 2026-04-25 — bump this whenever index rows, project status, or focus-window framing change. (Cosmetic edits to other pages don't require a bump here; log is the chronological ground truth.)" — to give blank-thread LLMs and human readers an immediate freshness signal at the top of the public-facing dashboard. Why option A and not B/C: B was the mkdocs-git-revision-date-localized-plugin (auto-derives "Last update" footer from git on every rendered page; would solve drift on all 9 pages with manual timestamps at once); C was B plus a content/metadata split (keep editorial dates like Snapshot 2026-04-24 on What-Im-Working-On, retire pure-metadata ones). Both deferred — option A keeps the dep surface small, ships in one minute, and the bump rule is narrow enough (only when index rows / status / focus-window framing change) that drift risk is low. Note also captured for future-Darcy: the manual-timestamp drift risk is real and the plugin is the right move eventually; revisit if a second pure-metadata timestamp gets added or if this one starts going stale. Not done: no mkdocs.yml plugin add; no requirements-pages.txt change; no edits to existing manual timestamps on other pages (they're all currently editorial-dating, not pure-metadata, so the split isn't needed yet).
  • Agentic Operating System frame folded in (harness-agnostic 7-layer vocabulary + chief-of-staff label + OS audit cadence): Triaged a podcast (NLW guest, Agentic Operating System, 2026-04) that names a portable 7-layer agent-OS shape (Identity / Context / Skills / Memory / Connections / Verifications / Automations) with a "chief of staff" orchestrator pattern. Decision: keep three surgical pieces; leave the rest on the cutting-room floor. Pulled in: (1) the 7-layer vocabulary as a cross-walk table on personal-operating-system (left column = portable layer, right column = current implementation here — soul.md/AGENTS.md/Grok-Brief for identity, wiki/ + raw/ for context, OpenClaw workspace skills + .cursor/skills-cursor/ for skills, OpenClaw memory hardening + ground-truth rule for memory, scoped Gmail/Calendar/Brave/Telegram/xurl for connections, Telegram gate + ground-truth rule + audit for verifications, Crusty daily routines + future Aye Robot loop for automations); (2) explicit chief-of-staff / Seat 0 label for projects/crusty in the Description block, with a one-line cross-ref into the existing Model-routing section explaining why the orchestrator is the expensive tier (this is the principled answer to that section's "counterintuitive but important" note); (3) a new OS audit cadence subsection on personal-operating-system tying an 8-week-shelf-life retrospective into the existing 2026-07-18 focus-window boundary — the agent-stack analogue of the goals quarterly review and the recurring-issues retrospective discipline. Why this lives on personal-operating-system specifically: the projects/openclaw-autonomy-org Hermes Agent analysis and the goals long-term thesis both already assume tools will swap; the layer vocabulary makes the wiki itself the OS-of-record so a Hermes seat / OpenClaude → Apple-OS-+-Gemini swap / future tool reshuffle changes only the right-hand column of the table, not the system. What was deliberately left out (logged for next time the same frame surfaces): the brain-dump / 15-question identity-interview methodology (it's how-to content, not project state — belongs in private notes if anywhere); specific skill catalog examples from the podcast (pre-read, daily brief, voice match, commitment tracker — Darcy's daily-routines table on projects/crusty is the right shape, listing what's actually run, not what was suggested); memory plumbing specifics (per the wiki-public-vs-private discipline already enforced after the 2026-04-20 security audit); connections least-privilege checklist (already implicit in Crusty's minimal-permissions posture and the Telegram-gate pattern, restating it would just add noise); automation safety rules (already covered by the ground-truth rule + draft-before-send default). Not updated: index (no new pages, no project-status changes — the personal-operating-system row + crusty row are unchanged in role); Grok-Brief (the 7-layer vocabulary is wiki-internal scaffolding, not cold-start context Grok needs in a blank thread); goals (Seat 0 / chief-of-staff name landed on projects/crusty where the orchestrator role actually lives, not on the goals page where it would compete with the Plausible fleet roster's own Seat 1–6 numbering — the chief-of-staff label is role, the seat numbers are positions, both can coexist); What-Im-Working-On (no near-term plan change — this is strategic-layer scaffolding, not a focus-window addition); mkdocs.yml (no nav additions — only existing pages were edited). No active-project status changes — Waypoint 1 + Waypoint 2 + Carlot conditional all unchanged; this is a vocabulary / framing layer that makes Crusty's existing role explicit so future-Darcy doesn't have to re-derive the harness-agnostic mapping every time a tool-swap question comes up.

2026-04-24

  • Creative Space (new) — dedicated studio / gym / build space, scoped as staged material goal: New projects/creative-space created from a same-day musing about a large converted-music-studio-style space supporting four activities on demand — projects/happy-body training + coaching, projects/rusty-cage / projects/music-production recording + playing, projects/the-gathering as candidate venue, and agent-building / business operation (including voice-pipeline QA for projects/buddyline and the voice channel on projects/education-startup-ux) — with real audio isolation and a workable vibe. Currently a recurring mental picture rather than a scoped plan; logging it makes it testable over time rather than re-costing attention every 6–12 months. Strategic framing explicitly captured: this is not a fleet-roster seat on goals — under the mission filter (push the fleet forward + strengthen one of context depth / trusted access / durable trust), a physical space is environmental infrastructure, not agent capability. Voice-pipeline QA is the only thin line connecting it to the thesis. Page sits alongside projects/the-gathering, projects/rusty-cage, projects/happy-body as life / identity infrastructure — legit on its own terms but competes with income-leg work for capital and attention, so needs an explicit status rather than permanent aspiration. Staged four-level plan documented: Level 0 (adapt current shared space, <$1k, one weekend — the only level active during the current focus window), Level 1 (convert outbuilding, $5–30k, partial reversibility), Level 2 (rent commercial / warehouse / studio space monthly, ~$300–900/mo PNW, often the right final answer), Level 3 (buy or build fully-converted studio — six figures, multi-year, explicitly deferred past 2027 and gated on a fleet income seat actually producing). Skipping levels is named as the main failure mode. Focus-window disposition (→ 2026-07-18): Level 0 only; Levels 1–3 explicitly not considered during the focus window because they would directly dilute Waypoints 1 + 2. Focus-window deliverable is a friction log (empty table on the page to populate whenever the shared-space constraint actually stops something, not just annoys) — gives the 2026-07-18 review real data rather than a vibes-only re-run of the same feeling. Explicit "honest first question" framing captured: fewer than ~6 friction-log entries by 2026-07-18 is strong evidence the constraint is aspirational rather than operational; ≥ 6 is the primary trigger for scoping Level 2. Other Level 2 triggers documented (a concrete Happy Body coaching engagement whose revenue covers rent; a Gathering chemistry signal wanting an owned-ish venue). Level 3 triggers documented as requiring all of (income seat on goals producing, Level 2 run ≥ 12 months with clear data, life situation supports multi-year commitment). Level 0 checklist sized to fit under $1k / one weekend (AI work surface, audio hygiene for voice QA, DI-only music practice path, dumbbell corner, consistent coaching-call framing + acoustic panels at first reflection points). Honest counter-arguments preserved on the page so 2026-07-18 Darcy rereads rather than re-derives: material goals often hide frustration rather than opportunity; committing Level-3 capital before the fleet produces income is the classic failure mode of lifestyle compounding eating thesis runway; shared-space constraint may be a feature (brake against 16-hour agent-coding spirals), not just a bug. Updated: index (new row in Life & identity sub-table), What-Im-Working-On (snapshot date bumped to 2026-04-24; new bullet under Context / other lines with the four-level summary and focus-window stance), mkdocs.yml (nav entry under Projects). Not updated: Grok-Brief (same pattern as projects/the-gathering + projects/buddyline + projects/carlot-chat intake — deferred life-surface candidates don't belong in the cold-start pack), goals (no fleet-roster change — explicitly not a seat), personal-operating-system (the space decision is the kind of system question that domain page would eventually hold, but the domain stubs are still TBD there; cross-ref added on the creative-space page rather than pre-filling the operating-system stubs with a half answer). No active-project status changes — this is a strategic-layer addition that names a long-standing material goal without pausing or reshuffling anything live.

2026-04-23

  • BuddyLine (new) — voice-first AI companion for seniors; Seat 5 reframed; voice-for-education folded into projects/education-startup-ux: New project projects/buddyline created from a same-day ingest (raw source at raw/buddyline-summary.md). Core concept: a phone-number-based AI companion seniors reach via their existing phone (no app, no download, no device), family-managed dashboard for setup, xAI low-cost TTS/STT + Grok brain, cron-style reminders, always-identifies-as-AI, explicitly stays out of regulated medical territory, positioned as a supplement to human contact rather than a replacement, no dedicated hardware. Strategic take captured in the page: BuddyLine is an unusually clean fit with the goals thesis — ✓✓ on all three legs (context depth, trusted access, durable trust), which is the cleanest score of any row in the alignment table. It is also the literal embodiment of the "agent, not app" philosophy underwriting projects/crusty, and has a real previously-unnamed synergy with projects/triviabalance (same senior audience; shareable content pipeline). Priority decision: explicitly not pulled into the current focus window — starting it now would make the window a four-line portfolio (Waypoints 1 + 2 + conditional projects/carlot-chat + BuddyLine) and scatter attention; the thesis fit is durable rather than time-sensitive; and the real infrastructure pre-reqs (telephony, STT/TTS, family dashboard, per-senior accounts) are the wrong order of operations while the Crusty model-routing decision is still open. Default revisit is 2026-07-18 at the quarterly goals review; earlier triggers named: (a) a concrete pull from a family member in Darcy's circle, (b) projects/triviabalance engagement data wanting a conversational surface, (c) an xAI voice-pipeline price/quality shift that changes the math. goals v1.4: Seat 5 reframed from "Consumer-product voice (conditional)" to "Consumer-product voice + senior companionship (conditional)" to cover both the original LunarCast-stance shape and the BuddyLine shape, with an explicit note that BuddyLine's materially higher trust surface (outbound calls, vulnerable demographic, real-time conversation) may justify splitting it into its own seat once prototyped. New BuddyLine row in the alignment table (✓✓ all three legs). projects/education-startup-ux: voice-for-education parallel experiment added as a new section ("Voice-first prototype direction") rather than a separate project page — it's a new channel on an existing product line; same xAI voice-pipeline bet, same agent-not-app philosophy, same safety posture adapted for very young kids (session transcripts available to parents, conservative content guardrails, no retention of children's voice data beyond session need); cross-references with BuddyLine since both live or die on the same voice pipeline and can share TTS/STT plumbing. projects/triviabalance: Related section and Known issues updated with the BuddyLine senior-audience synergy note and a design-hint that any new content pipeline should assume BuddyLine could consume it over voice. What-Im-Working-On: snapshot date bumped to 2026-04-23; new BuddyLine bullet added under Context / other lines making the deferred-but-strong-candidate status explicit. index: BuddyLine row added under Exploratory & background (the correct class given the deferred-but-active-idea status); triviabalance + education-startup-ux rows updated to surface the new cross-project links. mkdocs.yml: nav entry added. Not updated: Grok-Brief (same pattern as projects/the-gathering + the original projects/carlot-chat intake — don't pre-load the cold-start pack with deferred candidates; revisit if/when BuddyLine un-shelves). No active-project status changes — Waypoints 1 + 2 and the carlot-chat conditional are unchanged; this is a strategic-layer addition that names a strong candidate seat without pausing or reshuffling anything live.

2026-04-22

  • index restructured — Projects dashboard promoted to top of page, split into 5 priority classes: Reorganized wiki/index.md so the Projects surface leads the page (right after the Retrieval note), grouped into five scannable sub-tables: (1) Top priority — current focus window (→ 2026-07-18) = operator-agent (Waypoint 2), ayerobot-comic (Waypoint 1), crusty (infrastructure for both), carlot-chat (candidate pending confirmation). (2) Agent & context infrastructure = personal-projects-wiki, llm-maintained-context, openclaw-autonomy-org — the reusable substrate that everything else depends on. (3) Life & identity = rusty-cage, happy-body, music-production, personal-health-diet, the-gathering — recurring non-income rhythms. (4) Exploratory & background = triviabalance, education-startup-ux — active but explicitly outside the focus window. (5) Shipped & paused = lunarcast (live, growth paused), ai-model-distillation (Phases 2–4 paused). Moved Public / primary pages, Supporting pages, and Core maintenance below the Projects dashboard since those are maintenance metadata, not daily reading. Existing top-priority preface paragraphs (Waypoints framing, Model-routing decision, Claude-Code-on-Crusty TODO) preserved and relocated into the Top priority section where they belong. No content was removed — every project row from the prior flat table is accounted for in one of the five new sub-tables; only the ordering, grouping, and page placement changed. Rationale: the flat table was in historical-addition order, so a first-time reader had no visual signal about what was hot vs. paused, and the most-important surface was buried under five header sections. The new layout reads like a dashboard — what's active and urgent is first; what's shipped or paused is last.
  • Fleet roster — Seat 2b (productized-agent / AaaS surface) added + Carlot reclassified + The Gathering created: Strategic review landed three related updates on the public wiki. (1) goals v1.3: added Seat 2b — Productized-agent / AaaS surface to the Plausible fleet roster, sitting alongside Seat 2 (physical-node ops) rather than replacing it. Seat 2b owns "custom Crusty-style agents built for other small businesses — same draft → human-gate → execute → verify pattern applied to the client's surface" with medium blast radius (per-client credential isolation, never shared with Seat 1 or Seat 2). Added a "Two candidate income seats, not one" sequencing note below the roster making explicit that (a) Seat 2 and Seat 2b strengthen different sides of the thesis (Seat 2 = trusted access against real P&L; Seat 2b = context depth across multiple clients), (b) which one gets focus-window attention depends on which produces a confirmed engagement first, and (c) the long-term Option 4 synthesis — Seat 2b revenue funds Seat 2 entry at a higher rung than vending, shortening the path to the robotics endpoint. Alignment table gained a projects/carlot-chat row (✓ context depth, ✓ trusted access, ✓ durable trust — the first candidate engagement is already thesis-aligned). Meta bumped to v1.3. (2) projects/carlot-chat: status moved from "intake / scoping (side — outside focus window)" to "intake / scoping — candidate top priority pending engagement confirmation"; Type field now flags it as the first concrete Seat 2b candidate; rewrote the old "Why this is not in the 90-day focus window" section as "Focus-window status (revised 2026-04-22)" explaining the conditional top-priority framing; Known issues and Related cross-refs updated accordingly. (3) projects/the-gathering — new project page. Seeded as "recurring small-group forum for entrepreneurial and technically-curious peers; Seattle-area, in-person preferred; host-rotation and/or third-place venue; possible recorded / published format." Candidate formats documented (host-rotation at homes / shared third-place venue / hybrid with media), monthly weeknight cadence tentatively chosen, success criteria named (first gathering ≥ 4 attendees by end of 2026 Q3; group viability = ≥ 6 meetings with a stable core of ≥ 4 attendees over a year). Explicit non-goals called out for at least the first year: monetization, sponsorship, public brand building, group growth beyond one table. Seeding list itself deliberately kept in local-only notes (not in the public wiki). Framed as a people-graph surface alongside projects/rusty-cage and projects/happy-body, not a fleet-roster seat. What-Im-Working-On snapshot gained a "Candidate pivot (2026-04-22) — AaaS lane via projects/carlot-chat" bullet; context-lines section gained a one-liner for projects/the-gathering; pain points gained a Seat 2b engagement-confirmation bullet. index Carlot row updated with the conditional-top-priority framing; new the-gathering row added. mkdocs.yml gained a nav entry for The Gathering. Not updated: Grok-Brief (skipping until Seat 2b has a confirmed engagement to point at — no need to add a conditional to the cold-start pack yet). No active-project status changes on Waypoints 1 or 2; this is a strategic-layer addition that names a second income lane without pausing or killing anything that was already live. Rationale lives in private life-framing notes (gitignored, raw/); the public wiki carries only the neutral, project-framed version of the decision.

2026-04-21

  • The Car Lot — chat UX (consulting) intake page created: New projects/carlot-chat capturing a side / consulting engagement with a friend to rebuild the chatbot UX on https://www.thecarlot.ca/ (used-car dealer in Greater Sudbury, ON). Documented four key observations from public surfaces: (1) widget footer reads "Serviced by Gubagoo", so the incumbent is almost certainly Gubagoo (mainstream dealer chat platform, ~$300–700 CAD/mo band), not "Autodroid" as the owner recalled — autodroid.ai doesn't resolve as a live product, likely a half-remembered name, a reseller/rebrand, or a separate overlay; (2) Autoez CRM (autoez.ca/Inventory) looks bespoke (127 vehicles, 105 sync runs logged, inventory-sync job, Lead Activity + Delivered Sales tabs, eDealer Lead Export) — probably built by the friend and the right backend for the bot to query; (3) thecarlot.ca looks like a dealer-template site (possible vendors: EDealer / Strathcom / Convertus / DealerOn) — if vendor-hosted, we may only be allowed to inject a <script> tag, constraining every downstream design choice; (4) concrete UX failure captured in screenshot — bot was asked "What F-150s do you have in stock?" and responded "I'll be happy to have my team confirm… May I have your name?" despite F-150s being visible in the CRM, which is the exact "answer is one CRM query away" gap an AI answer layer fixes. Intake email to friend organized into five buckets: #1 usage (chats/day, conversion, after-hours share — single most important set, "improve UX" is unfalsifiable without it); #2 site ownership (domain + hosting, who built the site + CRM, DOM control, API access); #3 vendor + cost (confirm Gubagoo via invoice, monthly cost, contract length + renewal date, what owner likes/dislikes, 90 days of transcripts); #4 collaboration stack (shared repo for devs vs Basecamp/Notion/Google Doc for owner-facing thread — recommended: shared GitHub + shared doc for owner, Basecamp only if already in play); #5 optional design/scope (inventory/financing/trade-in/booking integrations, brand voice, EN/FR, mobile split, OMVIC disclaimer requirements). Three plausible project shapes documented (replace-the-widget / AI-layer-on-top-of-Gubagoo / bespoke-chat-on-a-separate-page) — pick after intake answers come back. Explicit non-focus-window framing: project stays in the wiki as single source of truth for intake + decisions, but must not compete with projects/operator-agent or projects/ayerobot-comic during the 90-day focus window (→ 2026-07-18); revisit scope then if it grows. Also flagged do not contact the vendor yet — contract-exit conversations happen between the owner and the vendor, not us. Updated: index (new Projects table row marked Intake / scoping (side — outside focus window) with cross-ref to the focus-window note), mkdocs.yml (nav entry under Projects, labeled "The Car Lot — chat UX (consulting)"). Not added to What-Im-Working-On or Grok-Brief — consulting side work should not appear in the top-priority LLM briefing surface while the 90-day focus window is active. No active-project status changes.
  • Hermes Agent — considered, deferred to post-2026-07-18 (file-of-record created on projects/openclaw-autonomy-org): Evaluated Hermes Agent (Nous Research, MIT, released Feb 2026, v0.8.0, ~42K GitHub stars) as a potential alternative to or complement for OpenClaw on Crusty. Alex Finn — the same YouTuber cited as prior art in the Model routing decision — has been promoting the "OpenClaw orchestrator + Hermes specialist" pattern; weighted his position with the caveat that he runs a Hermes bootcamp so the "use them together" advice has a sales-posture lens. Decision: do not add during the 90-day focus window. Three reasons, captured in full on projects/openclaw-autonomy-org (Hermes Agent (considered, deferred to post-2026-07-18)): (1) focus window already has three unblocked critical-path items (Claude Code TODO, Model routing, xurl install + live tweet) — none need Hermes; (2) Crusty's OpenClaw stack runs daily-reliable loops today and trading reliable-today-for-potentially-better-tomorrow is exactly the trade recurring-issues has caught us on before; (3) the gating question (does OpenClaw support per-skill model override in the current config schema? — same question as Model routing open handoff #1) isn't resolved, and the answer materially changes the Hermes case (if yes, ~80% of Alex Finn's pattern is achievable inside one OpenClaw config — Hermes's differentiator shrinks to "self-improving skills"; if no, two-process architecture exists regardless and the case for Hermes strengthens relative to "two OpenClaw sessions"). Weighted analysis captured: 3 pros (multi-agent composition is on the goals skills-to-deepen list = high weight post-focus-window; self-improving skill loop targets the confabulation failure mode = high-in-theory; Grok-only runtime invariant survives via OpenRouter = medium) vs 3 cons (complexity = high weight, and critically Claude Code mitigates debug-time but NOT design-time cognitive load — that tax is paid in "which framework owns which cron/skill/integration" decisions, not session debugging; self-improving skills in tension with the ground-truth rule = disqualifying for high-stakes orchestrator roles, fine for low-stakes; Alex Finn's firsthand note in his April 2026 OpenClaw tutorial that OpenClaw's memory is better than Hermes's in practice contradicts Hermes's strongest marketing claim = medium-high weight). Post-focus-window plan: if Waypoints 1 + 2 ship cleanly by 2026-07-18, try Hermes as a new fleet seat (not a Crusty reshuffle) — Seat 3 (wiki / context maintainer) is the strongest first-home because lowest blast radius + the self-improving skill loop is directly exercisable on the wiki itself with trivial audit. Seat 4-adjacent research-assistant probe for Operator Agent as a secondary option, read-only. Hard constraint captured: never Hermes as orchestrator for projects/ayerobot-comic or projects/operator-agent — that's the role being hardened with premium Grok, swapping it for a self-improving agent undoes the posture. Three revisit triggers named so future-Darcy doesn't have to re-derive: (a) Waypoints ship unusually cleanly + spare operational attention; (b) Model routing hits the "must split into two OpenClaw sessions" fallback (two-process architecture already paid for); (c) Hermes ships v1.0 (reduces breaking-change churn from the current v0.8.0 on a 4-month-old framework). Updated: projects/openclaw-autonomy-org (new ## section at end with Status, What it is + Alex Finn bootcamp caveat, Critical gating question, Weighted pros, Weighted cons, Decision, Post-2026-07-18 scoped experiment plan, Hard constraint, Revisit triggers), projects/crusty (one-line cross-ref in Related noting the Hermes decision lives on openclaw-autonomy-org as a fleet-seat question, not a Crusty-machine question), glossary (new Hermes Agent row). Not added to What-Im-Working-On — explicitly deferred work doesn't belong in active pain-points. No active-project status changes — this is a backlog decision getting the file-of-record it deserves so the analysis doesn't have to be re-run in three months.
  • Crusty — Claude Code as debug-time tool (TODO, force-multiplier decision added): Captured an open decision that's been hiding for months: up to now, Crusty debugging has happened from Darcy's main Mac because logging Cursor into Crusty would break the "dedicated, clean laptop, no personal-account linkage" posture (full IDE = claude.ai-style identity surface, codebase indexing to Cursor's infra, default-wired Anthropic/OpenAI endpoints). Alex Finn's pattern is to run Claude Code (Anthropic's CLI) on the OpenClaude box — Darcy conflated this with "Cursor" in conversation, but they're very different surfaces: Claude Code is a session-scoped CLI with no workspace indexing at rest, no IDE identity, credentials in macOS Keychain only. Reframed as a force-multiplier TODO, not a blocker: Crusty is the chokepoint for Aye Robot Waypoint 1, Operator Agent ops skills, daily routines, and the Model-routing decision itself, so seconds off every Crusty debug loop compounds across every active project. Two access options analyzed: (1) Anthropic Console API key — pay-per-use (Sonnet 4.6 $3/$15 per 1M in/out, Haiku 4.5 $1/$5), scope-restrictable via Claude Code role, no claude.ai identity, one-click rotate. Expected cost at Darcy's debug cadence: ~$2–8/mo. (2) Claude Pro subscription — $20/mo flat via claude.ai OAuth login, cannot "max out the sub" in a billing sense (Pro is flat, attacker ceiling is rate-limit DoS), but adds a full claude.ai identity to Crusty's graph and the OAuth scope isn't restrictable. Recommendation: start on API key path — matches Crusty's minimum-blast-radius posture, cheaper in expectation, zero web-login identity added. Invariant named and locked in: "Claude runs on Crusty, but Crusty doesn't run on Claude" — Claude Code is dev-time only, scoped to Darcy's interactive shell, never wired into OpenClaw skills; runtime autonomous loop stays 100% Grok (orchestrator + skills unchanged). Setup checklist included (9 steps, secret-touching steps flagged (manual — Darcy only)): fresh Console account on band identity, prepaid Visa, $10/mo hard spend cap, Claude Code-role API key, install via code.claude.com, paste key via first-run prompt to land in Keychain (not ~/.zshrc), pin Haiku 4.5 or Sonnet 4.6 via /model, /status verify, scope workspace to skill dir not repo root. Day-one Pro-path mitigations also captured in case the recommendation is reversed (band identity, history off, TOTP 2FA, two-step revocation). Sequencing: independent of the Model-routing decision; arguable best to install Claude Code first so the routing work itself is done with Claude Code on Crusty. Updated: projects/crusty (new Claude Code on Crusty (TODO — force multiplier, 2026-04-21) section with trade-off table, recommendation, 9-step setup checklist, Pro-path mitigations, open handoff questions, and sequencing notes; Known issues gets a top-of-list entry referencing the new section), What-Im-Working-On (new pain-points bullet leading with the force-multiplier framing + recommendation + invariant + cross-ref; stays above the Model-routing bullet since it's arguably the step-0 enabler), index (top-priority preface + Crusty status row flag the new TODO). No active-project status changes — this is a dev-tooling / posture decision; nothing live shifts until Darcy is back on the Crusty machine.
  • Mission added to goals (+ v1.2 fleet-roster restructure same day): Captured Darcy's explicit mission statement — "produce an autonomous fleet of agents that provide value and generate income 24/7" — as a new top-of-page Mission section on goals, above the existing three-legged North-star thesis. Named the distinction: the mission is what is being built (a fleet, plural — Crusty first with successors under projects/openclaw-autonomy-org, shared context layer, same draft → human-gate → execute → verify pattern everywhere); the thesis is why that bet compounds (agent era rewards context depth + trusted access + durable trust). Both are used together as the per-project filter — a project must push the fleet forward and strengthen at least one thesis leg, otherwise default to no. Explicitly framed income 24/7 as a real constraint (not just productivity theater) — this is why projects/operator-agent is a top-priority experiment rather than someday-maybe: it's the first test of whether the fleet can pay for itself via real P&L surfaces (supplier POs, Stripe reconciliation, customer support) rather than just automate work Darcy already does. Updated the Review cadence bullet to include the mission filter alongside the thesis legs. Second pass (v1.2, same day): restructured Positioning bets (5–10 year) from a flat list of strategic bets into a Plausible fleet roster table (seats 1–6, each with domain + current prototype + trust surface / blast radius) plus a short Non-seat bets list for outcomes that aren't agents — since the mission explicitly names a "fleet," the positioning section should read as who's in the fleet rather than a generic bet list. Roster: Seat 1 = Personal ops + creative (Crusty — live; daily routines + Aye Robot); Seat 2 = Physical-node ops specialist (starts as Crusty skill surface for projects/operator-agent node 1, likely splits into its own OpenClaw instance once node 2+ are live because Stripe = different trust bar from band inbox — this is the seat the income leg runs through); Seat 3 = Wiki / context maintainer (half-built today = Cursor + Darcy, lowest blast radius so a natural next autonomize); Seat 4 = Coaching / education surface (conditional, on-device preferred, minor-adjacent → highest trust bar); Seat 5 = Consumer-product voice (conditional, LunarCast stance as the reference); Seat 6 = Band / gig ops (optional, may stay a Crusty skill forever). Also named cross-seat invariants explicitly: shared wiki context (Seat 3's job), draft→gate→execute→verify, scoped secrets per seat, session-JSONL audit, identity stability over time. Non-seat bets kept as Pattern exporter, Trust-first product instincts, and Lived operator credibility (reputation earned by Seat 2 doing its job long enough, not a roster position). Meta bumped to v1.2 with both 2026-04-21 entries. Cross-references: index (row for goals now leads with the one-line mission), What-Im-Working-On (Direction paragraph prepends the mission ahead of the thesis; both link to goals), Grok-Brief (ultra-minimal paragraph now carries the mission line so blank-thread Grok gets it too). No active-project status changes — this is a strategic-layer addition that makes explicit what was already implicit across Crusty, Operator Agent, Aye Robot, OpenClaw org, and the wiki itself. Sequencing takeaway from the roster: Seats 1 + 2 are exactly the 90-day focus window's payoff; Seats 4–6 are explicitly conditional and should not compete for attention before 2026-07-18.
  • Crusty — model routing becomes top priority (gates Aye Robot Waypoint 1 resume): Named and captured an open decision that's been hiding in the project: Crusty currently runs a single model (grok-4-1-fast-reasoning) for everything — orchestrator, skills, heartbeats, digests, and the future ops surface — and reliability on complex judgment tasks isn't good enough for a live state-changing posting cadence. That's the exact failure class the "re-read public state before claiming success" rule on recurring-issues exists to catch, and also the exact concern the OpenClaw — tool-adapter class failure with reasoning-model function-call format entry's hedge predicted ("if a fast reasoning model regresses on complex tool use, try a fuller model for agentic work and keep the fast variant for lighter routines"). Decision: land per-role model routing before shipping the first live Aye Robot tweet — same pattern Alex Finn has productized on YouTube (orchestrator on premium reasoning; specialist / classifier / heartbeat workers on cheap models). Pricing confirmed (per 1M tokens at xAI's pricing page): grok-4-1-fast-reasoning is $0.20 / $0.05 cached / $0.50, vs grok-4.20-0309-reasoning at $2.00 / $0.20 cached / $6.00 — 10–12× per blended token. At current ~$2/mo Crusty spend, a naive full-stack swap to 4.20 projects ~$20–24/mo; per-role routing (premium orchestrator + ops, cheap everything else) projects ~$6–10/mo. Counterintuitive point locked in: the orchestrator is the expensive one, not the workers — it's where ground-truth / escalation / confabulation detection happens, which is exactly where Crusty has failed before. Open handoff questions (need Crusty-machine eyes): does OpenClaw support per-skill model override in the current config schema (if yes → one config-edit pass); if not → split Crusty into two OpenClaw sessions (orchestrator/ops on 4.20, routines on 4.1-fast, sharing .env + Telegram). Also: confirm prompt caching actually hits on Crusty's workloads before believing the $6–10 projection; confirm xAI credit rebate from linked teams is neutral at the new spend level; defer grok-4.20-multi-agent-0309 until plain 4.20-reasoning on the orchestrator proves insufficient. Sequencing: (1) resolve per-skill vs per-session on the Crusty machine; (2) flip orchestrator first in isolation, run a full dry-run Aye Robot pipeline; (3) move routine skills to cheap tier; (4) one day of live cost observation at console.x.ai; (5) only then proceed with xurl install/auth + first live tweet. Explicit prerequisite added as step 0 of the xurl handoff in projects/aye-robot-crusty-paused-x-automation so a future handoff cannot accidentally skip the routing change. Updated: projects/crusty (new Model routing (open decision — top priority) section with pricing table, tiering table, open handoff questions, sequencing, and why-routing-first-is-cheaper-than-debugging-later rationale; Description paragraph now flags the decision; Waypoint 1 gets an explicit prerequisite bullet; Known issues gets the routing item as top entry and marks Aye Robot posting as gated on it), projects/aye-robot-crusty-paused-x-automation (Status field flags resume blocker; Still open lists routing decision as hard prerequisite; xurl handoff prepended with step 0 "land routing first"), What-Im-Working-On (snapshot Waypoint 1 tagged with RESUME BLOCKER; Crusty blurb rewritten around the routing decision; pain points lists routing as #1; wins/momentum @ayerobotcomic entry rewritten as blocked on routing), index (new top-priority preface under Projects noting the routing-decision gate; Aye Robot status row and Crusty status row both flag the open decision).

  • Aye Robot / Crusty — next-step transport locked in as X's official xurl CLI: Folded in the Brian Roemmele handoff note and Darcy's confirmation that the band's X developer app already exists (Client ID + Secret, Pay-per-use package, Production environment — both required or API v2 calls return client-forbidden). Decision: the steady-state posting transport is X's official xurl CLI (github.com/xdevplatform/xurl) sitting behind OpenClaw's already-merged skills/xurl/SKILL.md — not the custom post_comic_test.py OAuth 1.0a script (that was bring-up only; it proved credentials/policy/pricing shape in dry-run, now demoted to reference). Why xurl: official CLI from X; OpenClaw auto-recognizes the merged skill (natural-language surface for xurl post, xurl reply, xurl search, xurl timeline, xurl whoami, etc.); OAuth 2.0 PKCE handled by xurl auth oauth2; native media upload so no custom wrapper; stays 100% local on Crusty; no third-party bridges; pay-per-use pricing unchanged (~$1.20/mo). Added a new "2026-04-21: Next steps — switch transport to xurl" section to projects/aye-robot-crusty-paused-x-automation with the six-step handoff (install via brew install --cask xdevplatform/tap/xurl; xurl auth apps add + xurl auth oauth2 run manually by Darcy — not the agent, since secrets are involved; smoke-test sequence xurl whoami → xurl timeline -n 3 → xurl post; OpenClaw restart; updated daily Aye Robot prompt that chains grok-imagine → Telegram approval → xurl post --media). Open questions re-scoped: dev-app provisioning closed (already exists); v2 media-upload pricing remains open as pure cost-model hygiene (mechanism is now xurl's problem, not ours). Updated projects/aye-robot-crusty-paused-x-automation (Status, Still open, new xurl next-steps section, closed open-question, verified-and-working script note), projects/ayerobot-comic (Status, Description, target workflow Post/Listen rows, sequencing, Known issues transport contract), projects/crusty (Aye Robot paragraph, daily routines, Waypoint 1 pattern, known issues), projects/operator-agent (Waypoint 1 line), projects/openclaw-autonomy-org (Aye Robot bullet, marketing-target bullet), What-Im-Working-On (snapshot, wins, pain points, Crusty blurb), index (status table row + waypoint line), recurring-issues (Aye Robot status bullet, new X dev-app config trap bullet for client-forbidden on non–Pay-per-use / non-Production apps, Auth-model trap bullet rewritten to surface OAuth 2.0 PKCE as the xurl path alongside OAuth 1.0a), glossary (Aye Robot row), Grok-Brief (ultra-minimal paragraph + Waypoint 1 clause + Crusty / Aye Robot / Aye Robot comic capsules + Operator Agent 90-day note), tools-and-repos (new xurl entry under Email / CLI). No project-status changes — "active / E2E unblocked" still correct; just the steady-state posting script identity is now xurl, not post_comic_test.py.

  • Aye Robot / Crusty — unpause status (E2E unblocked on host): Documented 2026-04-21 state on projects/aye-robot-crusty-paused-x-automation: grok-imagine workspace skill + xAI grok-imagine-image working; post_comic_test.py X API v2 path working (dry-run; live tweet not yet sent). OpenClaw / config fixes captured in an ordered table (valid JSON, no top-level x_api, no bad grok-imagine-image under models.providers, no imageGenerationModel pointing at xAI, SKILL.md paths + single-command exec rule, TPM raise at console.x.ai, .env fallback for X keys in poster, repaired comic.json, fixed projects.md newline serialization, workspace docs so Crusty uses grok-imagine not image_generate). Still open: scheduling (no cron/launchd yet), live smoke-test post, optional Telegram approval wiring, v2 media-upload pricing / dev-app checks if needed. Synced: projects/ayerobot-comic, projects/crusty, What-Im-Working-On, index, recurring-issues, projects/openclaw-autonomy-org, glossary, Grok-Brief (ultra-minimal + capsules).

2026-04-20

  • Aye Robot / Crusty unpause — three corrections for reality drift: Fixed three specifics in projects/aye-robot-crusty-paused-x-automation (and the supporting rule in recurring-issues) where the 2026-04-18 writeup was accurate at the time but drifted from what's actually on Crusty / what X's API now documents. (1) Media upload = v2, not v1.1. Old wording pointed at the v1.1 /media/upload path; corrected to POST /2/media/upload with explicit chunked routes POST /2/media/upload/initialize → .../:id/append → .../:id/finalize. v1.1 media upload still responds but the docs index now lists it as legacy, so all new work targets v2. Open question #1 on media-upload pricing re-scoped to the v2 endpoints. (2) Auth model = OAuth 1.0a user-context (four keys), not Bearer Token. Old step 2 said "load the X API bearer token" — this conflated app-only Bearer (works for read-mostly endpoints) with user-context auth (required by POST /2/tweets). Bearer tokens cannot call POST /2/tweets; posting needs the four OAuth 1.0a user-context keys (API Key, API Key Secret, Access Token, Access Token Secret) or OAuth 2.0 Auth Code + PKCE. Step 2 rewritten to spell this out, with OAuth 1.0a named as the pragmatic choice for a single-handle daily loop (PKCE's refresh-token dance isn't worth it for one owner). Added an explicit "Auth-model trap for X API v2" entry on recurring-issues so a future 401 with an otherwise-valid Bearer doesn't send anyone hunting for a credential bug that's actually an auth-model bug. Referenced post_comic_test.py from this handoff as the reference OAuth 1.0a implementation. (3) xAI key lives in .env, not the OpenClaw config file (aspirational-vs-actual resolution). The wiki said in two places that runtime keys "load from the OpenClaw config file (not .env / profile files)" and that the X API key should follow "the same pattern as the existing xAI key" — confirmed with Darcy 2026-04-20 that reality is the opposite: the xAI key has always been in Crusty's .env, with the OpenClaw config referencing it via ${env:VAR} expansion. Two options to close the drift were considered: (a) migrate xAI key into the config file to match the wiki's stated pattern, or (b) update the wiki to match reality. Picked (b) — the port handoff's post_comic_test.py already uses the .env + ${env:VAR} pattern, and rewiring xAI mid-unpause is unnecessary risk. Canonical pattern now documented: one .env is the single secret source; the OpenClaw config is the wiring layer that names which env var feeds which skill; shell profile files (.zshrc / .bash_profile) stay out of it entirely. Incoming X API OAuth 1.0a keys follow the same pattern. Updated: projects/aye-robot-crusty-paused-x-automation (line 18 xAI key note, unblock step 2, unblock step 3, open question #1), recurring-issues (old "key-loading precedence" clause removed; new Key-sourcing pattern (actual, confirmed 2026-04-20) bullet + Auth-model trap for X API v2 bullet). No active-project status changes — the port plan, pricing math, and Telegram-gate architecture remain correct; only the implementation specifics were off.
  • Long-term goals page (new): New goals — strategic compass anchored on a Peter Diamandis quote ("the winners of the next decade won't be the apps with the best UIs; they'll be the agents with the most trusted access to your data — the agent era rewards context depth and trust"). Darcy agrees and is betting on it. Page is deliberately one level above What-Im-Working-On: what's here is the thesis ("why the current projects are the current projects"), what's there is the short-range plan. Core structure: (1) three-legged thesis — context depth, trusted access, durable trust; (2) why it fits Darcy specifically — projects/crusty + projects/personal-projects-wiki + projects/operator-agent were already implementing this pattern before the thesis was named, plus temperament fit (low-blast-radius instincts), leverage (context + trust are scarce skills vs. prompt engineering), and physical+digital combo is rare; (3) alignment table — every active project mapped against the three legs (every active project scores at least one ✓; pure-UI / growth-hack plays correctly score zero and are already paused or killed); (4) skills to deepen — agent trust architecture (draft→gate→execute→verify), personal context engineering (wiki-as-memory), agent ops / reliability, on-device + distilled models, physical-world agent ops, multi-agent composition; (5) positioning bets (5–10 yr) — personal agent operator, operator of agent-run physical nodes, pattern exporter (SCHEMA + wiki pattern), trust-first consumer instincts (LunarCast stance), optional education/mentoring surface; (6) explicit anti-thesis filter — pure UI plays, growth-hacking, dropship / content farms, LLM wrappers with no context layer, crypto / franchise / pure consulting, platform-dependent automation (the old consumer-web X path is literally an example of this, conveniently); (7) quarterly review cadence tied to the 90-day focus windows. Cross-references added: index (new row in Public / primary table), What-Im-Working-On (one-sentence thesis added to Direction paragraph + step 2 of LLM how-to), Grok-Brief (one-sentence thesis appended to the ultra-minimal paragraph so blank-thread Grok gets it too), mkdocs.yml (new top-level nav entry "Goals (long-term)"). No active-project status changes — this is a strategic layer only.
  • Security audit + public-site hardening + repo-privacy flip: Full audit of what the wiki exposes now that it's intended as public LLM context. Outcome: (1) GitHub repo flipped to private (GitHub Pro supports public Pages from a private repo, so wiki.darcymenard.com stays public while the source tree — including git history going forward, raw/, .github/workflows/, mkdocs.yml, editor state — stops being world-readable). (2) Public rendered pages (wiki/) scrubbed of Crusty targeting-packet detail: file paths (~/.openclaw/, ~/.crusty/), gateway port, config-key names, OpenClaw version numbers, BotFather rotation instructions, hardware specifics, specific model IDs. Replaced with generic phrasing ("dedicated, clean laptop"; "xAI / Grok-only stack"; "session-log audit trail"). Touched projects/crusty (rewritten to a public-safe skeleton, full operational detail moved to private notes), projects/aye-robot-crusty-paused-x-automation, recurring-issues, projects/ayerobot-comic, projects/openclaw-autonomy-org, projects/lunarcast, What-Im-Working-On, Grok-Brief, index, glossary, SCHEMA.md. (3) Public-minor de-identification on projects/happy-body and Grok-Brief — "daughter" → "two people" for the mentoring reference; same for related passages. (4) mkdocs.yml: removed edit_uri (kills the "Edit this page" prompt-injection-via-PR vector for LLM-consumed pages). (5) .github/workflows/pages.yml: pinned actions/checkout, actions/setup-python, peaceiris/actions-gh-pages to commit SHAs as supply-chain defense-in-depth. (6) .gitignore + untracks: .obsidian/ editor state stopped being tracked; raw/ was briefly gitignored pre-privacy-flip but the privacy flip makes that unnecessary — re-tracking raw/ is a follow-up. Caveat: anything previously committed while the repo was public is reachable in historical clones and GitHub's cached commit views; the privacy flip stops future exposure, not past. For the small number of items that truly matter (the PHI line, the minor's identification, the specific cold-outreach targeting packet in raw/operator-agent/node-1-model.md), consider a git filter-repo + force-push history rewrite if the past-exposure tail risk is high enough to justify it.
  • Aye Robot / X policy ingest (simplifies project): Re-read X automation rules (October 2025 update) and folded the policy into projects/aye-robot-crusty-paused-x-automation (new Policy posture table + updates to Where it stalled, If you resume later, and the 2026-04-18 unpause section). Net simplifications: (1) browser automation is permanently off the table — "Don't use non-API-based automation like scripting the X website" is a named permanent-suspension offense, so the consumer-web browser-automation path moves from "paused/maybe revisit" to a hard no; the local sandbox stays installed for non-X web recon only. (2) Closed open question #3 on the unpause plan — "is 1 image/day 'normal human account behavior'?" — X's own rules list "automatically broadcast helpful information in posts" as an allowed use case, so cadence is explicitly compliant, not gray-zone. (3) Telegram gate gets a stronger 'why': "account holders are ultimately responsible for automated activity" means the gate is policy-aligned, not just defensive design — same code, stronger justification. (4) Added minor future guardrails for off-X matching (needs opt-in) and redistribution (use Post IDs, not full content) in case the archive/analytics scope ever expands. Also touched projects/ayerobot-comic (Known issues) and recurring-issues (Aye Robot next-step bullet). No changes to the port plan itself — API v2 transport + preserved Telegram gate remains correct and is now explicitly endorsed by X policy.
  • Wiki context boundary (you-are-not-Crusty callout): Added an explicit "the LLM reading this wiki is not Crusty" disclaimer in three places so future agents (Cursor, Grok, anything pointed at the public site) don't confuse the wiki-maintainer context (Darcy's main personal machine) with the Crusty context (separate dedicated laptop, minimal permissions, no access to this account/repo/personal accounts; this machine in turn has no access to Crusty's inbox / bot / model-provider keys / local config). Updated index (retrieval note), projects/crusty (new "Note to LLMs reading this wiki" section above Description), SCHEMA.md (first bullet under "Tips specific to this project"). No substantive project-status changes.

2026-04-18

  • Aye Robot unpause + two-waypoint framing (late-day revision): Elevated projects/ayerobot-comic autonomous loop back into the active focus window, reframed alongside projects/operator-agent as two waypoints on a single learning curve — "Crusty runs a real automated workflow end-to-end with a Telegram human-gate, reliably, day after day" — rather than two competing priorities. Waypoint 1 = Aye Robot (quicker win, low stakes); Waypoint 2 = Operator Agent (advanced version of the same pattern, real P&L). The architectural pattern (draft state-changing action → Telegram gate → execute → verify via public state / session JSONL) is identical across both; only the integration endpoints and stakes differ. The Aye Robot port is explicitly positioned as de-risking the operator ops prototype: prove the gate pattern under boring weekly pressure before Crusty touches real money. Trigger: X announced pay-per-use API pricing on 2026-04-17 (effective 2026-04-20) — owned reads $0.001/req, text posts $0.015, summoned replies $0.01, URL-in-post $0.20 (13×), auto-like/auto-follow/quote-post removed from self-serve; no subscription tier (X API pricing). This specifically resolves the economic + transport blockers documented in projects/aye-robot-crusty-paused-x-automation — at current cadence, full autonomous loop runs ~$1.20/mo. Scope of Waypoint 1: port X posting transport from Playwright+Brave → X API v2 (POST /2/tweets + media upload); Telegram approval gate preserved; time-boxed to 2 weekends (by ~2026-05-03) — if the port slips past that, re-triage the four open questions in projects/aye-robot-crusty-paused-x-automation rather than grind further. What stays paused: projects/lunarcast growth push (its tactics — auto-like/follow/quote-post — were killed in the same pricing change), projects/triviabalance, distillation Phases 2–4, education-startup side work. Out of scope for the unpause: growth engineering, quote-tweet threads, any engagement-farming pattern. Note: the framing was revised same day after an initial "two competing priorities with scatter-risk guardrails" writeup; the "two waypoints on one curve" frame is the correct one and is now primary in all docs. Updated projects/aye-robot-crusty-paused-x-automation (Status field + unblock section), projects/ayerobot-comic (Status + workflow + known issues), projects/crusty (Direction rewritten with waypoint framing + daily routines), projects/operator-agent (Focus commitment with waypoint framing + Known risks reframed as scope-creep risk + Related), What-Im-Working-On (snapshot + direction + pain points + wins), index (status table + header waypoint note), Grok-Brief (ultra-minimal paragraph + Operator Agent + Crusty + Aye Robot + Aye Robot comic capsules + phrase routing + shortcuts), mkdocs.yml (nav label).
  • Operator Agent (new): New projects/operator-agent — physical-node business experiment; node 1 = smart vending / micro-market (PNW, Seattle-area); long-term path documented: EV charging (L2) → laundromat / car wash acquisition → robotics-as-a-service → own hardware / robotics build; 2-week feasibility sprint (2026-04-18 → 2026-05-02) with go/no-go gate; 90-day focus commitment (→ 2026-07-18). Crusty re-targeted from marketing (blocked on X web automation — projects/aye-robot-crusty-paused-x-automation) to ops agent (telemetry, supplier POs, Stripe reconciliation, support). Paused for 90 days: projects/ayerobot-comic autonomous loop, projects/lunarcast growth push, projects/triviabalance, projects/ai-model-distillation Phases 2–4. Kept as identity/life: projects/rusty-cage, projects/happy-body. Updated projects/crusty (Direction re-pointed at operator ops), What-Im-Working-On, index, Grok-Brief (routing + capsule), mkdocs.yml nav.
  • Aye Robot / Crusty: New projects/aye-robot-crusty-paused-x-automation — paused (2026) autonomous daily comic loop; verified: workspace/skills, image-generation skill, session-log exec audit, OpenClaw config as key source of truth, Telegram, local sandbox + Playwright/browser smoke; stalled: unattended X via consumer web. Updated projects/ayerobot-comic, projects/crusty, recurring-issues (session-log vs gateway log, key precedence), projects/openclaw-autonomy-org, What-Im-Working-On, index, Grok-Brief, mkdocs.yml nav.
  • Crusty: OpenClaw stack is xAI / Grok only on this instance (no Anthropic/OpenAI API). Marketing target: autonomous projects/ayerobot-comic loop + lighter projects/lunarcast promo; easiest path on projects/crusty. Touched Grok-Brief, What-Im-Working-On, projects/openclaw-autonomy-org, projects/ayerobot-comic, projects/lunarcast, glossary, recurring-issues, index.

2026-04-17

2026-04-15

  • Personal operating system: New personal-operating-system — constraints → delegation template; Marketing & self-promotion (Crusty, planned marketing-playbook.md); stub domains for interview follow-up. index (supporting pages), What-Im-Working-On (LLM how-to step2), mkdocs.yml nav.

2026-04-14

2026-04-12

2026-04-11