agentleFS
ucsandmanucsandman/DashClaw/.claude/skills/dashclaw-weekly/SKILL.md

dashclaw-weekly

Skill310 starsChanged 6 months agoMITView on GitHub

Compile the week's maintainer-log entries + git history into ONE pasteable Reddit-ready DashClaw update. Drafts only — publishing stays human (MAINTAINER.md §4 spirit). Use when Wes asks for the weekly update, a Reddit digest, or "what happened this week".

  • Reads credentials

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Log entries: read the covered entries in docs/maintainer-log.md — they are the primary source (decisions, incidents, numbers).

    line 18
  2. Distinctive:

    Draft only: Never post, never call outbound services.

    line 52
  3. Very specific:

    After Wes confirms he posted it, append <!-- digest-posted: YYYY-MM-DD --> on its own line at the TOP of the entries section of …

    line 56

≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

DashClaw weekly digest

Turn the maintainer's log into a public weekly update Wes can paste to Reddit unmodified. The story is the experiment: an AI maintains this repo under a human-held constitution (MAINTAINER.md), and these are its receipts.

Gather (in order)

  1. Cursor: find the last <!-- digest-posted: YYYY-MM-DD --> marker in docs/maintainer-log.md. Cover everything after it. No marker → last 7 days. An explicit range in the user's args overrides both.
  2. Log entries: read the covered entries in docs/maintainer-log.md — they are the primary source (decisions, incidents, numbers).
  3. Commits: git log --oneline --no-merges --since=<cursor> on main — cross-check nothing shipped unlogged. If something did, say so in the digest (the gap is itself content) and append a catch-up log entry.
  4. Dogfood stats (best-effort, skip silently if unavailable): with a local instance running and DASHCLAW_API_KEY in .env.local, GET /api/policies/summary (30-day decision outcome counts — label the window honestly) and /api/policies/proposals (open tuning proposals). These are the "the product governs its own maintainer" numbers.

Compose

One fenced code block, ready to paste. No hard wraps — one line per paragraph, blank lines between (Wes's standing formatting rule for copyable text). Plain language; define DashClaw terms on first use; no internal codenames, file paths only when they're the point.

Sections, in order, dropping any that are empty this week:

  • Title: DashClaw maintainer update — week of <date> (+ version if one shipped)
  • TL;DR: 2-3 sentences.
  • Shipped: what landed, with version + one-line why-it-matters each.
  • Decisions: the why behind non-obvious calls, from the log.
  • Incident of the week: the most instructive failure, told straight — these are the credibility of the whole series. Never omit one that's in the log.
  • The AI is governed by the product: dogfood stats when available.
  • Next up: the next roadmap item, one line.
  • Links: repo, MAINTAINER.md, docs/plans/owner-roadmap.md, docs/maintainer-log.md.

Hard rules

  • Draft only. Never post, never call outbound services. Deliver the block in chat; Wes publishes.
  • Log is source of truth — do not soften incidents or inflate numbers; every figure must trace to the log, git, or a live API response.
  • After Wes confirms he posted it, append <!-- digest-posted: YYYY-MM-DD --> on its own line at the TOP of the entries section of docs/maintainer-log.md (above the newest covered entry) so the next run starts after it. Not before confirmation.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

Reviews can't be read right now.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.