agentleFS
kapadiaskapadias/nonna/.claude/skills/fast-lane/SKILL.md

fast-lane

Skill0 starsChanged 47 days agoMITView on GitHub

Deciding whether a change qualifies for the bounded fast lane (~<=15 lines, <=3 files, no new deps, off the critical surface). Used by /fix.

The file

Files
SKILL.md

Fast lane — proportionality with a deterministic fence

The full loop (Research → Plan → TDD → Implement → Review → Verify → Ship → Sync) is the default for a reason: it is where mistakes get caught. But a one-character typo fix and a schema migration are not the same risk, and forcing both through the identical eight stages teaches people to bypass the loop instead of trusting it. This is the autonomy dial: adjustable process depth, with the adjustment made by a script, never by self-assessment. An agent that could argue its own change into the fast lane would put the fox in charge of the henhouse — so eligibility is decided by scripts/check-trivial.sh, which fails closed.

Eligibility — the script decides

bash .claude/skills/fast-lane/scripts/check-trivial.sh [base] exits 0 only if all hold:

  • ≤ 15 changed lines across ≤ 3 files (added + deleted, vs the merge-base with develop by default). Test/fixture paths and docs/STATUS.md do not count against the budget — a fast-lane fix should be mostly test.
  • No dependency manifest or lockfile touched — a new dependency is never trivial (supply-chain skill).
  • No critical-surface path touched: .claude/hooks/**, .claude/settings.json, .claude/skills/*/scripts/**, .github/workflows/**, **/migrations/**. Extend per-repo via NONNA_CRITICAL_PATHS (colon-separated globs) — anything touching money, auth, persistence, or irreversible actions belongs on it.
  • No ambiguity: not-a-repo, unresolvable base, or a binary change all exit 1. When classification is uncertain, the full loop applies — fail closed, per rules/safety.md.

The budgets are constants at the top of the script. Raising them is an explicit, reviewable act.

What the fast lane skips — and what it never skips

Skipped (proportionality)Never skipped (non-negotiable)
The written plan (/plan)The failing-test-first rule (rules/testing.md)
Orchestrator fan-outThe full local test gate
Two-reviewer parallel /reviewEvery hook (branch, secret, DoD)
ADR ceremonyThe machine review verdict (check-review.sh)
The one-line docs/STATUS.md entry

A single code-reviewer pass replaces the two-reviewer fan-out; its JSON verdict still goes through check-review.sh with the identical severity gate. Behavior changes still pin a regression test first (RED → GREEN); only docs/comment-only diffs — which have no behavior to pin — skip it.

If the script says no

Route through the full loop (/plan → /tdd → /review → /ship). Do not argue with the classifier, shrink a change artificially to sneak under the budget, or split one logical change into several "trivial" pushes — salami-slicing past a gate is a bypass, and the reviewer is told to flag it.

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.