agentleFS
talas9talas9/anti-hall/.claude/skills/engine-lane/SKILL.md

engine-lane

Skill2 starsChanged 46 days agoMITView on GitHub

Dev-only, anti-hall repo (not part of the shipped plugin). Use when running or briefing an ah-engine (Rust) lane - isolated clone, queued builds, targeted tests, the per-train full gate and replay, the no-hardcoding gates, Node goldens, and cleanup.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 2 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

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

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

    Every agent or lane brief names its issue #n.

    line 15
  2. Distinctive:

    Before starting, check free disk and load.

    line 29
  3. Very specific:

    Every cargo call goes through the shared build queue, from ah-engine/: AH_CARGO_PRIO=<0-9> ~/.anti-hall/work/bin/cargo-q.sh <cargo args> …

    line 33
  4. Very specific:

    Every threshold, limit, interval, text, phrase list, rule, setting or dispatch row lives in the plugin's config …

    line 49
  5. Distinctive:

    Never delete user data.

    line 80
  6. Very specific:

    After the patch is handed over (commits fetched or a patch file written), delete the lane's ah-engine/target/ (multi-GB) and remove the …

    line 86

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 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

engine-lane: running an ah-engine lane

A lane is one bounded engine change (one issue, see gh-work) built in isolation and handed back as commits for the coordinator to integrate. Every lane brief restates these rules. A machine may keep its own stricter lane rules (for example ~/.anti-hall/work/LANE-RULES.md); read them first, they win over this summary.

HARD RULE: every agent and lane is tracked on its issue

  • Every agent or lane brief names its issue #n.
  • At start: move the item to In progress and comment the approach.
  • At milestones and at the end: comment evidence (commit SHAs, test names and results).
  • A paused or stopped lane gets a state comment immediately.
  • PRs use Closes #n.

1. Where to work

  • A fresh scratch clone or a git worktree add under ~/.anti-hall/work/ (never /tmp, never the shared checkout other sessions use). One mutating lane per checkout.
  • Base it on the branch named in the brief (engine work branches off the engine branch, not main). To test an older commit use git worktree add <dir> <sha>; never git stash.
  • Git identity = the repo's git config; never --author, never set user.email; no AI-credit lines. Lanes never push unless the brief says so; never force push.
  • Before starting, check free disk and load; lanes never outnumber build slots.

2. Building and testing

  • Every cargo call goes through the shared build queue, from ah-engine/: AH_CARGO_PRIO=<0-9> ~/.anti-hall/work/bin/cargo-q.sh <cargo args> (priority FIFO, bounded slots, niced, disk-guarded). Never bare cargo, never cargo run.
  • Lane = targeted tests only: cargo-q.sh check, then only the tests of the modules you touched, e.g. cargo-q.sh nextest run -E 'test(<module>)'. Not the parity/integration sweeps, not the full suite, not a replay.
  • Once per release train (on the merged train branch, not per lane): the full gate (ah-engine/test.sh: clippy -D warnings, the whole nextest suite, timed tests alone, doctests, no surviving daemon), the soak profile, and the frozen replay (the Node goldens replayed against the engine; any check weaker than Node is a P0). A train failure is bisected over its lane merges. CI fixes are batched: reproduce the failing job locally, push once.
  • Tests never touch the real HOME (~/.claude, ~/.anti-hall): scratch HOME, fixture repos.
  • No pkill/killall; kill only PIDs your lane started. No background polling loops.

3. No hardcoding, nothing compiled (D88)

  • Every threshold, limit, interval, text, phrase list, rule, setting or dispatch row lives in the plugin's config (plugins/anti-hall/engine/defaults/ plus its byte-identical defaults.pristine/ twin), read at start-up, on plugin update and on file change. Decision logic belongs in plugin JavaScript (plugins/anti-hall/engine/logic/<check>.js, run by the embedded interpreter), not in compiled Rust.
  • Gates that must pass:
    • no_hardcoded_tunables (cargo-q.sh nextest run --release -E 'test(no_hardcoded_tunables)'): no tunable literal in engine source.
    • compiled_logic_gate (tests/compiled_logic_gate.rs against tests/compiled_logic_ceiling.txt): the count of non-scripted checks may only go down; lower the ceiling when you script one.
  • True constants (protocol or format versions) only through the narrow, justified allowlist.
  • New verbs, features and settings go in the engine registry (so the generated engine skill picks them up) and declare which roles may use them.

4. Node parity and goldens

  • Zero new Node. New capability only in the engine; shipped Node gets bug fixes only and stays in parity. The engine is never weaker than Node: anything not provably identical defers (nothing written).
  • ah-engine/tests/goldens/<lane>/ holds recorded Node answers. A change to a Node file in a lane's fingerprint fails with stale Node golden <lane>: <file> changed; re-record and commit the golden with the Node change: AH_RECORD_NODE=1 cargo-q.sh nextest run --release --profile record -E 'test(<test>)', then review git diff tests/goldens/<lane> (usually only manifest hashes). Recording is refused under CI=true. Details: ah-engine/tests/goldens/README.md.

5. Acting features

New features act by default; each action emits a telemetry event (feature, action, target, inputs, outcome, reason, latency, action id) and later mistake signals, reported by the engine's telemetry/metrics verbs. Keep idempotency, a live re-check, an in-flight lock and the ledger. Never delete user data.

6. Hand-off and cleanup

  • Report: branch, commit SHAs, exact test commands and counts, open issues. Comment the same evidence on the lane's issue.
  • After the patch is handed over (commits fetched or a patch file written), delete the lane's ah-engine/target/ (multi-GB) and remove the scratch worktree once the coordinator has the commits (git worktree remove <dir>; see repo-hygiene).

Model routing for agents and lanes: see the "Model routing" section in .claude/skills/README.md.

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.