---
name: <FILL: skill-name>
description: <FILL: one-line — what this autonomous mode explores/produces, and when it's the default action>
license: MIT
tier: any
category: intelligence
autoInvoked: false
dependencies: []
relatedSkills: [cos, checkpoint, ship]
requirements:
  env: []
  integrations: []
---

<!--
  AUTHORING A NEW AUTONOMOUS MODE
  ==============================
  This is the meta-template for a NEW domain-specific autonomous skill (e.g. a
  sibling to /bethesda and /wallstreet). The shared 7-phase loop, claim-before-
  work, capture obligation, hard-stops, and self-evolution all live in
  docs/claude/autonomous-mode-contract.md — DO NOT re-specify them here. You only
  fill five domain blanks + your domain lens, then add domain-specific phase
  detail where the contract leaves a blank.

  Replace every <FILL: ...> marker, delete this comment, delete unused scaffolding.
  Exemplars to copy patterns from: .claude/commands/bethesda.md, wallstreet.md.
-->

# /<FILL: skill-name> — <FILL: tagline>

<FILL: 1-2 sentences of orientation — what world this skill operates in.>

> **Autonomous Mode.** This skill inherits the invariant loop, **claim-before-work**, the
> capture obligation, hard-stops, and self-evolution from
> [`docs/claude/autonomous-mode-contract.md`](../../docs/claude/autonomous-mode-contract.md).
> It fills five domain blanks:
> - **SOURCES** — <FILL: what to read at orient — which channels / MCPs / vault paths / feeds>
> - **WORK-UNIT / CLAIM-KIND** — a *<FILL: cave|message|artifact|quest|…>*; keys `<FILL: cave:<id> | ticker:<SYM> | artifact:<slug> | daily:<skill>:<date> | beat:<slot>>`.
> - **SCORING** — <FILL: how to pick the next card from the candidate set>
> - **CAPTURE-FORMAT** — <FILL: what the TraqrDB memory + diary entry look like for this domain>
> - **APPROVAL-GATE** — <FILL: usually "none; ships via Guardian (daemon-pr)" — only name a gate if this domain genuinely has one>
>
> Authority order on conflict: **CLAUDE.md > the contract > this skill.** Claim a card with
> the shared CLI **before** working it (Phase 0, claim step below).

## North Star (read every session)

<FILL: 2-4 bullets — which of your project's strategic vectors this serves, the decision-rubric default
(decide-and-ship), and the one domain-specific principle that should never be forgotten.
Reference CLAUDE.md for the rubric/vectors; don't restate them.>

## Phase 0 — Clock in (orient + read the board)

1. <FILL: load this skill's SOURCES — the orient reads. Reuse `scripts/bethesda-orient.sh`
   (or the generalized orient) for the shared context block; add your domain feeds.>
2. **Read the claim board** (advisory — narrows candidates):
   ```
   npx tsx packages/coordination/src/cli.ts list
   ```

## Phase 1 — Claim a card (BEFORE any work)

Once a candidate is obvious, **claim it atomically before investigating, editing, or building:**
```
npx tsx packages/coordination/src/cli.ts claim <key> --by=<slot> --ttl=1800 --note="<card>"
```
Derive `<key>` by kind (see the contract's key table; use the deterministic helpers, never raw
strings). **Exit 0 = you hold it, proceed. Exit 2 = already held → pick the next candidate and
claim that** (never work a card you don't hold). Invoke via direct `tsx` (not `npm -w`, which
adds cosmetic stderr noise on the held-path). Then announce TAKING as a *post-claim notification
carrying the key*. Renew on each `/loop` heartbeat if work outlasts the TTL.

## Phase 2 — Do the work

<FILL: the domain-specific execution pattern — the actual cave/research/action shape. This is
where most of your skill's unique value lives. Keep the orient→capture→cascade scaffolding lean;
the contract owns it.>

## Phase 3 — Capture (proof of work, non-negotiable)

<FILL: the exact TraqrDB memory shape + diary entry for this domain (your CAPTURE-FORMAT blank).
A future agent must be productive in 30 seconds from the memory.>

## Phase 4 — Cascade or stand down

Pick the next card (claim it) or stand down. Hard-stops (rate limit / the principal steers / P1 /
saturated field) live in the contract — don't re-specify.

## Phase 5 — Reflect (optional)

At most one `[autonomous-improvement-proposal type=<friction|surprise|subtraction>]` capture,
only if a real one surfaced. Silence on an ordinary session is correct.

## Phase 6 — Clock out

**Release the card** (`… release <key> --by=<slot>`), wrap, reset dashboard phase. Skip the
wrap post if `/ship` already posted.