agentleFS
jblosseyjblossey/houserules/AGENTS.md

houserules

AGENTS.md2 starsChanged 20 days agoMITView on GitHub

This repository develops the kit in template/ and runs its own installed copy of it (entry houserules.template-is-the-source). Project knowledge — rules, invariants, gotchas, procedures, decisions, history — lives in knowledge/*.json. Every entry has an id. Every instruction above holds for any agent or harness reading this file, the dispatch mandate included: no instruction here depends on subagents existing, but none of them becomes optional for a harness that has them. One directory gives Claude Code a genuinely optional acceleration of…

  • Installs packages

The file

We split this AGENTS.md 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. Distinctive:

    Do not add knowledge to this file.

    line 14
  2. Very specific:

    Run houserules standing at the start of a session, at every batch start or resume, and before any dispatch.

    line 18
  3. Very specific:

    Before you change a file, run houserules for <path> and read what comes back.

    line 19
  4. Distinctive:

    Every instruction above holds for any agent or harness reading this file, the dispatch mandate included: no instruction here depends on …

    line 27
  5. Very specific:

    Write houserules output to a file and read the file.

    line 68
  6. Very specific:

    Run houserules for <path> before you change a file that matches.

    line 79

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.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
AGENTS.md

houserules — the kit repository

This repository develops the kit in template/ and runs its own installed copy of it (entry houserules.template-is-the-source).

Knowledge base

Project knowledge — rules, invariants, gotchas, procedures, decisions, history — lives in knowledge/*.json. Every entry has an id.

  • houserules topics lists the topics. houserules index --topic <t> lists a topic's entries.
  • houserules get <id>... prints full entries. houserules for <path>... prints the rules for the files you are about to change.
  • houserules standing prints the non-negotiables.
  • houserules check-knowledge and houserules check-backlog are lint gates. houserules audit --base <ref> checks a change against its rule package.
  • Write every ruling to its home file in the same turn it is made: scope or product goes to backlog/ (houserules set ... or a direct edit); process goes to knowledge/process.json (standing: true when non-negotiable); design goes to the batch spec. Then run houserules render and commit the generated files with the change.
  • Do not add knowledge to this file. Add an entry.

Workflow

  • Run houserules standing at the start of a session, at every batch start or resume, and before any dispatch.
  • Before you change a file, run houserules for <path> and read what comes back.
  • The backlog (backlog/) drives all work. Read it with houserules list --open, get <id>, batch <n>; tick with set <id> status=done batch=<n>.
  • A batch runs: brainstorm or spec, a user gate, plan, development, live run, a merge, rollout, acceptance.
  • Procedure docs live at .claude/skills/*/SKILL.md — plain markdown any agent can read; read .claude/skills/finishing-a-feature/SKILL.md before a merge.
  • Dispatch subagents only through the templates in .claude/agents/: implementer, task-reviewer, branch-reviewer, spec-reviewer. Every dispatch carries Knowledge: ids and BASE:.

Working with other harnesses

Every instruction above holds for any agent or harness reading this file, the dispatch mandate included: no instruction here depends on subagents existing, but none of them becomes optional for a harness that has them.

One directory gives Claude Code a genuinely optional acceleration of the same rules, never a requirement:

  • .claude/rules/*.md loads into Claude Code's context when you read a matching file: the rules of one area. The standing rules reach every harness through the houserules block of this file; Claude Code reads it through CLAUDE.md. Other harnesses get the rules of an area from the AGENTS.md of its directory or from houserules for <path>.
<!-- houserules:begin -->

Generated from knowledge/ by houserules render. Do not edit.

Standing rules

  • [houserules.breaking-changes-need-an-owner-ruling] A change that would prompt a major release (a ! commit or a BREAKING CHANGE footer) needs an owner ruling before it merges.
  • [houserules.tag-pilot-is-read-only] Treat the tag-pilot checkout outside this repository as read-only: never modify it, commit there, run its agents, or run init or update against it.
  • [houserules.template-is-the-source] Edit the kit in template/, then run houserules update --dir .; never hand-edit the root copies or the generated rules and skill.
  • [process.ask-when-missing] Ask the user when information is missing. Do not assume.
  • [process.claims-match-artifacts] Before submitting a report, re-open every artifact a claim cites and confirm the artifact shows what the sentence says.
  • [process.closure-claims-carry-enumeration] A claim that a set is closed ('every', 'all', 'no remaining') names the bounded, rerunnable enumeration that produced it; without the artifact, no claim.
  • [process.conventional-commits] Commits use Conventional Commits (feat, fix, chore, test, ci, docs, refactor). Header at most 100 characters, body lines at most 100.
  • [process.evidence-outlives-the-session] Cite evidence only at paths that outlive the session: the project's deliverables folder or the tracked tree, never a session scratchpad.
  • [process.fix-proofs-extend-the-reviewers-run] A fix round's proof re-runs the review's own reproduction, extended to cover the fix -- never a fresh, narrower proof.
  • [process.gates-cover-generated-commits] Before a new gate or pipeline step ships, walk every commit author class and every generated resource it touches through it on paper; state each accepted red.
  • [process.no-tech-debt] Fix every review finding, Minor included. Defer a fix only for a stated reason, as a work item in the tracker; never as a TODO in code.
  • [process.review-findings-are-claims-too] A review finding stating a measurement, a reproduction, or a proposed fix's behavior is verified by running it before filing; restated claims are re-derived.
  • [process.rulings-to-file] Every ruling goes to its home file in the same turn; one that changes what the spec states amends the spec; one inside its intent goes to the brief or report.
  • [process.suggestions-are-unverified] Another agent's suggested wording, pin, or count is a draft, not evidence: re-derive every number, quote, and reference from the tree before transcribing it.
  • [process.tdd] Test-driven development for every executable change: the failing test first, or a disclosed-mutation proof for already-correct behavior; verbatim RED and GREEN.
  • [quality.no-compat-softening] Never soften a design for backward compatibility. Make the correct change and migrate everything it breaks; the codebase is not built on compromises.
  • [quality.principles] YAGNI, KISS, DRY; prefer a well-maintained library to custom code; one responsibility per unit; tests assert behavior; a11y and i18n are part of done.
  • [quality.well-maintained-libraries] Use well-maintained libraries over custom code. Research candidates at spec time; the owner rules the choice; refactors fold into specs.
  • [security-hygiene.dependency-vetting] Verify every new dependency is well-maintained before adding it. Record the check in the task report's dependency_vetting field.
  • [security-hygiene.exact-pins] Pin exact versions. Install only with a CLI (cargo add <crate>@=<version>, pnpm add --save-exact). Never write a version number by hand.
  • [security-hygiene.no-coauthor] Never add a co-author, session, or tool-attribution line to any commit or pull request description.
  • [security-hygiene.verify-current-docs] Ground code, specs, and plans on the current docs of each library, tool, and framework and on the current knowledge entries. Internal knowledge is stale.
  • [writing-style.code-comments] Write a code comment only for a current constraint the code cannot show. Never narrate history in code; git, the specs, and the knowledge base hold it.
  • [writing-style.doc-comments] Document every exported symbol with the language's doc-comment convention: the current contract. Name things so the code reads without comments.
  • [writing-style.principles] Write docs, comments, commit messages, and reports in ASD-STE100 style: short sentences, active voice, one instruction per sentence, concise.

Session

  • Start: run houserules list --open.
  • End: run houserules check-backlog.
  • Write houserules output to a file and read the file; treat a << marker or a "more items" note as a lost read.

Reference workflow

This install runs the reference workflow. Read .claude/skills/orchestrating/SKILL.md at the start of a session and before any handoff. If your harness can hand a task to another agent the way .claude/agents/*.md describes, run handoffs in parallel only when the plan marks their file sets disjoint, each in its own worktree; a harness with no subagents does the work directly, in its own session.

Area rules

docs

Globs: docs/**, README.md, AGENTS.md, knowledge/**, backlog/**, .claude/** Run houserules for <path> before you change a file that matches.

tools

Globs: tools/**, .github/**, mise.toml Run houserules for <path> before you change a file that matches.

cli

Globs: crates/houserules/src/** Run houserules for <path> before you change a file that matches.

template

Globs: template/** Run houserules for <path> before you change a file that matches.

tests

Globs: tests/** Run houserules for <path> before you change a file that matches.

rust

Globs: crates/**, Cargo.toml, Cargo.lock Run houserules for <path> before you change a file that matches.

<!-- houserules:end -->

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.