agentleFS
talas9talas9/anti-hall/.claude/skills/dogfood/SKILL.md

dogfood

Skill2 starsChanged 46 days agoMITView on GitHub

Dev-only, anti-hall repo (not part of the shipped plugin). Use in every anti-hall development session. The session that builds anti-hall also runs it - observe every anti-hall message, telemetry, engine health and bug signal, collect reports from other projects' running sessions, log every misfire with evidence, and drive fixes.

81 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use dep-refs, feature-boundary-planner and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • axiom-desuSkill

    dep-refs

    Use when needs to inspect TypeScript export references in the z-code workspace, list exports from a file, verify whether an export is unused before deletion, investigate who imports a symbol during refactors, or combine pnpm knip unused-export results with pnpm dep:refs symbol-level reference tracing.

    Also in 6 projects that use this

  • axiom-desuSkill

    feature-boundary-planner

    Map a ZCode behavior change to current UI surfaces, state owners, protocol commands, persistence, and validation. Use for impact analysis, product-boundary planning, or an implementation handoff grounded in the checked-out source.

    Also in 6 projects that use this

  • lir-navcooSkill

    yuanbao

    Yuanbao (元宝) groups: @mention users, query info/members.

    Also in 18 projects that use this

  • Cartograf666Skill

    architecture-governance

    Bootstrap, document, enforce, and review a project's architecture through layer responsibilities, dependency direction, architecture manifests, legacy baselines, Change Envelopes, cross-layer approval policy, and deterministic architecture checks. Use when starting a project, auditing an existing codebase, planning cross-layer work, preventing architecture erosion, reviewing structural changes, or deciding whether implementation may expand into another layer.

    Also in 6 projects that use this

  • math-incSkill

    minecraft-modpack-server

    Set up a modded Minecraft server from a CurseForge/Modrinth server pack zip. Covers NeoForge/Forge install, Java version, JVM tuning, firewall, LAN config, backups, and launch scripts.

    Also in 5 projects that use this

  • math-incSkill

    pokemon-player

    Play Pokemon games autonomously via headless emulation. Starts a game server, reads structured game state from RAM, makes strategic decisions, and sends button inputs — all from the terminal.

    Also in 5 projects that use this

  • +1 more from math-inc

The file

We split this skill into the rules it gives an agent

2 rules. 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 81 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Every progress report to the owner includes: open entries by severity, fixed since last report, mistake rates of acting features.

    line 53
  2. Distinctive:

    Never close or delete an issue automatically.

    line 98

“Common” is measured: kept by at least 6 in 10 of the 81 edited versions of this file we could read. “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

Dogfood anti-hall while developing it

Dev-only: this skill lives in the repo's .claude/skills/, outside plugins/anti-hall/, so it is never installed for users (like a devDependency). Every anti-hall signal in a dev session is test output. Judge it, record it, fix it. Never silently obey, dismiss or work around one.

1. Every anti-hall message, as it arrives

Blocks, advisories, nudges, tracker lines, stop-hook feedback, statusline segments, DevSwarm notices. Verdict: true positive / false positive (wrong about facts) / noise (right but useless or repeated) / wrong wording. Anything but a true positive gets one line in .anti-hall/history/dogfood/ISSUES.md (local, gitignored):

- YYYY-MM-DD HH:MM | <check / event / source> | <verdict> | <message quote + contradicting fact> | open

A repeat of an open entry gets its own dated line (frequency is data). Keep working.

2. Telemetry and health (hourly, at every milestone, and before ending the session)

  • sh ~/.anti-hall/ah-engine-live/status.sh: errors, panics, restarts, breaker, crash-loop, daemon RSS.
  • sh ~/.anti-hall/ah-node-shadow/node-shadow.sh --compare: any check where the engine is weaker than Node is a P0 entry; defer counts per check are tracked over time.
  • ah-engine devswarm ledger --since 1h (and ah-engine telemetry): per-feature actions, success rate, refusals, mistake rate. A rising mistake rate or a new refusal reason is an entry.
  • The engine event log: whole-event fallbacks, stuck workers, script "interrupted"/CPU-limit errors, git timeouts. Record count and check id.
  • The repo itself: a stale .git/index.lock, temp-dir growth (ls $TMPDIR | wc -l), leftover ah-engine serve processes, disk growth. Each is a bug signal until proven otherwise.

3. Other projects' running sessions (every few hours and at milestones)

anti-hall runs in every project on this machine. List live sessions (ListAgents, peer sessions with status busy/idle, not offline) and send each a short message asking for anti-hall misfires, wrong blocks, slowdowns or confusing messages they have hit since the last ask, with the exact message text and what they were doing. Ask; never instruct them to change anything. Record each answer (or "no reply") as entries with source peer:<session name>. Owner-ratified peer channel; keep it to one message per session per round. DevSwarm workspace sessions are never messaged directly (devswarm-comms-guard blocks it, correctly): ask that project's Primary to collect from its workspaces and relay.

4. Fixing

  • Group open entries per check into one lane: confirm the cause from code/data, sibling sweep, regression test proving the true-positive case still fires, then the fix (plugin config/JS or engine; Node gets bug fixes only and stays in parity).
  • Mark entries fixed <sha> / wontfix <reason>; never delete entries.
  • Every progress report to the owner includes: open entries by severity, fixed since last report, mistake rates of acting features.

5. All work lives on GitHub (issues + the "anti-hall roadmap" board)

Tasks, bugs, features, plans and the roadmap live in the repo's issues and on the user project board "anti-hall roadmap" (project 3, owner talas9; views Now, Next, Backlog, By milestone, Board). The session task list is built FROM the board, never the other way round. Exact commands, field ids and option ids: the gh-work skill.

Setup this relies on

  • Issue forms (.github/ISSUE_TEMPLATE/): Bug report, False positive, Feature request (blank issues off). Their Priority / Estimate / Area / "Estimate in hours" fields feed the automation.
  • Labels: type:feature|bug|docs|chore, priority:P0..P3, size:S|M|L|XL, area:* (engine, devswarm, hooks, statusline, codex, ci, docs, release, plus the guard areas), status:triage|accepted|in-progress|blocked.
  • triage.yml: maps form fields to priority:*/size:*/area:*, adds status:triage on open, labels PRs by path (labeler.yml), and adds new issues to the board when its token secret is set (else the board's own auto-add). community.yml: rules triage + one model brief comment. The moderation workflows (when present) add moderation:review, needs-issue, risk:* and triage suggestions: a session acts on them (see gh-work).
  • Milestones: v0.300.0 (engine decides hooks, Node is the fallback), v0.301 engine-only (Node removed), v1.0 (no Node; everything goes through the engine; contract frozen).

Operating loop (autonomous)

  1. Session start / resume: sync the board (gh project item-list 3 --owner talas9), read the Now view, then Next by priority. Each session task subject starts with #n.
  2. Anything new (an owner request, idea, bug or plan, or something found while working): open the issue FIRST, from the form fields, and triage it at once (labels, board fields, milestone). Plans go in the issue body or a linked comment; a local plan file links its issue.
  3. Pick the next work yourself: the highest-priority Now item with no in-flight agent, else the top of Next. Move it to In progress (status:in-progress + board Status), comment the approach, work it, comment the evidence, and close it through a PR with Closes #n. Never wait for permission on accepted items. Stop only at owner-only gates (billing, secrets, destructive actions): comment on the issue, label it status:blocked, take the next item.
  4. Every few hours (with the health round of section 2): re-sync the board, triage anything new (bot suggestions included), post a one-line progress comment on each in-progress issue, and handle status:in-progress issues silent for 24 h and status:triage issues older than a day.

Rules

  • Triage = status:accepted + priority/size/milestone, or a comment saying why it is deferred. Never close or delete an issue automatically.
  • One comment per milestone (approach, lane/commit shas, test or replay evidence, blockers), not per step.
  • Branch <type>/<n>-slug, commits reference #n, PR body Closes #n, so a merge moves the item to Done.
  • Done: the closing comment states what was verified and how (command and result). Work shipped without proof stays open with the missing check named.
  • Dogfood entries that need a code fix get an issue (type:bug, area of the check), linked from the ISSUES.md line (issue #n).
  • Issue text stays agnostic: no private paths, names or session ids (public repo).

Rules that still apply

Evidence before verdicts. Never disable a guard to get past it; log it and work within it. Safety switches change only with the owner's OK.

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.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

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.