agentleFS
sequant-iosequant-io/sequant/CLAUDE.md

sequant

CLAUDE.md1 starsChanged 3 months agoMITView on GitHub

The file

We split this CLAUDE.md into the rules it gives an agent

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

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    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:

    Do NOT add Co-Authored-By lines to any commits in this repository.

    line 5
  2. Very specific:

    When invoking a sequant skill via Skill(skill: "<name>", ...) from inside another sequant skill, qualify names that collide with Anthropic …

    line 17

≈ “Common” is an estimate: at least 1 in 5 of the 26,980 CLAUDE.md 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
CLAUDE.md

Sequant

Commit Rules

  • Do NOT add Co-Authored-By lines to any commits in this repository.

Branch protection

  • main rejects direct pushes (ruleset CC, #1109: required test + canary, strict up-to-date — GH013; the payload in scripts/ruleset-main.sh swaps strict for a merge queue, which applies once #1368 lands — the live ruleset is unchanged until then). Every commit, docs included, lands through a PR; /release Step 6 does the same (#1131).

Hooks

  • HOOK_BLOCKED: Force push — see .claude/skills/_shared/references/force-push.md for the user-handoff pattern. Do not attempt CLAUDE_HOOKS_DISABLED=true bypasses; they don't work.

Skills

  • When invoking a sequant skill via Skill(skill: "<name>", ...) from inside another sequant skill, qualify names that collide with Anthropic top-level skills as Skill(skill: "sequant:<name>", ...). Bare colliding names silently misroute to Anthropic's version. Enforced in CI by npm run lint:skill-calls (scripts/lint-skill-calls.ts). See #562 / #568.

Testing

  • Gate tests ship with a recorded mutation result. A test whose job is to gate a claim — a fixture exists, a skill section is present, a flag is wired — must be mutation-verified before it counts as coverage: delete the thing it asserts, confirm exactly that test fails, restore, and record the result using the Mutation-verified: AC-N — <mutation>; <test> failed; restored. marker format (machine-anchored as <!-- SEQUANT_MUTATION: {"ac":"AC-N","mutation":"...","failedTest":"<test file path> > <describe> > <case>"} -->) in the PR body — /qa §6i parses and enforces this record for gate-test ACs, and it resolves only the path segment before the first >: a describe-block name or bare test title there classifies test_not_in_diff and floors the verdict at AC_NOT_MET. Scope file-reading assertions to the delimited region they mean to check; matching the whole file lets a doc header or comment satisfy the assertion. See #830, where deleting the injection fixture's payload left the suite green, and #939, which promoted this record from prose to a parseable gate.

Red main

  • A failing push run on main is reverted first and debugged second. When CI is red on main, the first change that lands is the revert of the commit that broke it — not a forward fix, not an investigation branch, not a "I'll have a patch in ten minutes". The revert PR references the failing run by URL, and the investigation happens afterwards on a branch off a green main. Rationale: every hour main is red, every other PR's CI is red too, so "pre-existing failure, unrelated to this diff" becomes true for everyone and the merge gate stops meaning anything — which is exactly how #1086 went a week unfiled while #1036, #1042 and #1048 shipped past it. The main ruleset (scripts/ruleset-main.sh --print, applied per docs/internal/graph-2026-09-guard/ruleset.md) requires the test check (strict up-to-date today; a merge queue once #1368 applies the #1350 payload), so a red main also blocks every merge until the revert lands. There is no field anywhere — PR body, issue comment, or template — for excusing a failure instead. Open the revert as a draft and re-run the failed job in parallel (gh run rerun <id> --failed); a green re-run closes the revert unmerged, a red one marks it ready and it lands. The commit hook rejects the revert: type — use chore(#N): revert ….

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.