Personal operating system¶
How constraints (what I’m bad at, allergic to, or shouldn’t spend time on) map to systems (delegation, agents, playbooks, rituals) so work still happens without fighting my nature.
Per domain, capture:
| Field | Meaning |
|---|---|
| Shortcoming | Honest limit: skill gap, energy pattern, or values conflict. |
| System | Who/what runs it, where the charter lives, cadence, and guardrails. |
| Goal | What “good” looks like for you, not generic success. |
Related: projects/crusty (OpenClaw delegate), Grok-Brief (human + Grok routing). Marketing charter (planned): wiki/marketing-playbook.md — not created yet; Crusty delegation references it once it exists.
Marketing & self-promotion¶
Shortcoming: Low natural skill at promotion; aversion to anything that feels pushy or cheesy.
System: Full delegation to Crusty (OpenClaw agent). See wiki/marketing-playbook.md (to be written) for charter, guardrails (80/20 value-first, organic only), daily tasks, and approval workflow.
Goal: Authentic, useful amplification without me becoming “the marketing guy.”
Portable layers (harness-agnostic)¶
OpenClaw / Cursor / Claude Code / Hermes / whatever-comes-next all converge on the same seven-layer shape — different file names, same job. This table is the cross-walk so the system survives any harness swap without rewriting the wiki. Frame credit: an Agentic Operating System podcast episode (2026-04, NLW guest) that names the layers in tool-agnostic language; this section is the projection of that frame onto Darcy's actual stack.
| Portable layer | What it is | Current implementation here |
|---|---|---|
| 1. Identity | Who the agent works for, communication style, hard rules. | OpenClaw soul.md on projects/crusty; Cursor AGENTS.md + rules under .cursor/; Grok-Brief as the cold-start identity capsule for blank Grok threads. |
| 2. Context | A curated, dated, on-demand library of your situation (priorities, stakeholders, ops principles). | The wiki/ tree itself — index, goals, project pages. raw/ is the immutable input layer (gitignored, not published). The raw → wiki pattern on SCHEMA.md is the context-curation discipline. |
| 3. Skills | Reusable instruction sets ("when X → do Y using sources Z → produce format F"). | OpenClaw workspace skills on the Crusty machine (e.g. grok-imagine, merged xurl/SKILL.md); Cursor skills under .cursor/skills-cursor/; per-project playbooks (e.g. the planned marketing-playbook.md referenced above). |
| 4. Memory | What persists across sessions vs. what's discarded. | Tool-level: OpenClaw memory hardening on Crusty (compaction flush + cross-session search) — config keys live in private notes only. Wiki-level: this repo is shared long-term memory for any LLM pointed at it. Discipline: the "re-read public state before claiming success" rule on recurring-issues. |
| 5. Connections | How the agent reaches real systems. | Crusty (2026-05-10): Telegram approval channel + X API via xurl + web search + scoped shell for morning Aye Robot; band-mail / meal digests on grok.com, not OpenClaw (detail: projects/crusty). Posture: read-only first, write only after observed reliability — same principle the projects/openclaw-autonomy-org minimal-permissions stance already enforces. |
| 6. Verifications | The lightweight checks before output is trusted. | Telegram approval gate for state-changing actions (projects/ayerobot-comic / projects/operator-agent); ground-truth re-read rule on recurring-issues; periodic OS audit (see below). |
| 7. Automations | Things the agent runs unattended. | Crusty: morning Aye Robot only (2026-05-10). grok.com: rustycagetrio mail summary + meal ideas. Rule: only automate after manual reliability is proven; outputs default to drafts before they default to sends. |
Local inference (main Mac — not Crusty): For some sensitive or local-only material Darcy wants off typical cloud chat/API paths, LM Studio on the main personal Mac runs a quantized MLX build of Google Gemma 4 (instruction-tuned; 4-bit in current use — 16 GB / M2 Pro 2023-class machine). No overlap with Crusty’s Hermes / Grok credentials or runtime; this is a separate trust boundary choice. “Local model” means weights + inference stay on-device relative to cloud LLMs — not a guarantee against app telemetry, screenshots, or paste errors.
Why this lives here, not on a tool page: the projects/openclaw-autonomy-org Hermes Agent analysis and the long-term goals thesis both assume tools will swap. Naming the layers in harness-agnostic language means the system-of-record is this wiki, not any one tool's config. If OpenClaw gives way to Hermes (or Apple OS-level + Gemini, or whatever the 2027 swap looks like), only the right-hand column changes.
projects/crusty is Seat 0 / chief-of-staff. Across the seven layers, Crusty is the orchestrator — the chief-of-staff seat that fans work out to specialist skills, owns escalation and audit, and (counterintuitively — see the Model routing section) runs the expensive model precisely because that's where ground-truth judgment lives. New fleet seats on goals inherit identity, context, skills, memory, connections, and verifications from this layer rather than re-deriving them — that's the compounding return that justifies hardening Crusty before standing up Seat 2, Seat 2b, or Seat 3.
OS audit cadence¶
The system has a roughly 8-week shelf life if not maintained — context files go stale, skills stop firing, automations drift from their original specs, identity rules accumulate without review. Discipline:
- At each focus-window boundary (currently 2026-07-18): audit which OpenClaw skills actually fired, which
wiki/context files haven't been touched since the last review, which automations failed silently, which identity rules need updating against new failure modes captured on recurring-issues since last time. - Between boundaries: every time you catch yourself re-explaining situation context to a fresh LLM thread, that thing was missing from the context layer — add it to the relevant project page or to Grok-Brief in the same pass.
The OS audit is the agent-stack analogue of the goals quarterly review and the recurring-issues retrospective discipline — the same compounding move applied to the operating system itself rather than to any one project.
Domains to fill (stubs)¶
Use the same three fields for each. Remove this section once populated.
Calendar, scheduling & logistics¶
- Shortcoming: TBD
- System: TBD
- Goal: TBD
Email & correspondence¶
- Shortcoming: TBD
- System: TBD
- Goal: TBD
Deep work & context switching¶
- Shortcoming: TBD
- System: TBD
- Goal: TBD
Finances & admin¶
- Shortcoming: TBD
- System: TBD
- Goal: TBD
Health, food, training¶
- Shortcoming: TBD
- System: TBD
- Goal: TBD
Creative output (music, apps, writing)¶
- Shortcoming: Easy to treat music as “when I have time” instead of protected recovery time.
- System: Rusty Cage on a ~two-gig/month + biweekly rehearsal cadence; harmonica most days or every other day — 1st-position foundation, cross + slant (2nd / 3rd) for musicality (major + minor on cross), 12th + overblows later — slow reps for breath and muscle memory; blues / singing as therapy; producer lane documented on projects/music-production alongside apps (e.g. projects/lunarcast) and wiki/docs.
- Goal: Stress relief and skill compounding (voice, breath, cardio spillover) without turning every hobby into content obligation — how-to video idea stays optional because the market is crowded.
Meta¶
- Last updated: 2026-05-17 (Local inference: main Mac LM Studio + Gemma 4 MLX 4-bit for off-cloud-path text; not Crusty. Prior: 2026-05-05 Creative output.)
- Interview / expansion: Answers from the constraint-and-delegation interview (see chat) should replace
TBDstubs and add/remove domains as needed.