agentleFS
oeveryoevery/rsp/AGENTS.md

rsp

AGENTS.md1 starsChanged 8 days agoMITView on GitHub

RSP tracks current work, stable specs, and archives under .rsp/. Use supplied or confirmed paths; discover only missing paths or applicable ancestor instructions, without overlapping whole-tree scans: 1. Nearest AGENTS.md for project or module instructions. 2. Root CONTEXT.md if present, then the relevant nearest CONTEXT.md for domain language, relationships, and navigation. If a legacy CONTEXT-MAP.md remains, retain its relevant context and use Core's context-migration branch; discovery alone does not authorize migration. 3. Establish the request's scope, authority and existing owner.…

  • Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.

    Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.

  • In this repository10 other files it gives its agents.

The file

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

6 rules. About 0 are common, 3 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Use supplied or confirmed paths.

    line 8
  2. Very specific:

    Do not treat .rsp/specs/ or .rsp/changes/ as replacements for nearest AGENTS.md or CONTEXT.md.

    line 15
  3. Distinctive:

    Keep tests focused on observable behavior and public command output rather than duplicated generated content or Skill prose fragments.

    line 27
  4. Very specific:

    Default rsp-doc owns repository-document, Skill and release writing through conditional artifact methods.

    line 33
  5. Distinctive:

    Treat host metadata limits as hard constraints, but do not use Skill body or reference word-count ceilings as correctness gates.

    line 35
  6. Very specific:

    Keep overlapping global engineering workflow Skills disabled in the maintainer environment while dogfooding so RSP capability gaps remain …

    line 36

≈ “Common” is an estimate: at least 1 in 5 of the 41,581 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

rsp

<!-- rsp:begin -->

RSP Entry

RSP tracks current work, stable specs, and archives under .rsp/.

Use supplied or confirmed paths; discover only missing paths or applicable ancestor instructions, without overlapping whole-tree scans:

  1. Nearest AGENTS.md for project or module instructions.
  2. Root CONTEXT.md if present, then the relevant nearest CONTEXT.md for domain language, relationships, and navigation. If a legacy CONTEXT-MAP.md remains, retain its relevant context and use Core's context-migration branch; discovery alone does not authorize migration.
  3. Establish the request's scope, authority and existing owner. A clear bounded request selecting one capability with sufficient owner facts loads that project Skill directly from .agents/skills/, without a Core or Focus preflight. For tracked work read the selected Change; for grouped work read its sibling Group Brief first. Read only relevant Specs and Decision Records under the configured authoritative path.
  4. Load rsp at .agents/skills/rsp/SKILL.md for unresolved routing or ownership, actual cross-capability coordination, or lifecycle operations. Read .rsp/focus.d/ only when needed to resolve or recover the owner; marker paths are open-work candidates and optional bounded Markdown is recovery guidance, not authority. An explicit WorkRef or the user's mentioned Change selects first; otherwise use current status and dependency evidence.

Direct capability invocation does not make tracked work untracked. Preserve existing Changes, required tracking prerequisites, Group dependencies, independent review and permission boundaries; missing prerequisites stop dependent work. Skills are the only workflow: hosts may discover them or read them directly, but a missing or unusable required Skill stops dependent work and must be reported as an incomplete installation. Do not use retired Rules or install implicitly. Do not treat .rsp/specs/ or .rsp/changes/ as replacements for nearest AGENTS.md or CONTEXT.md.

This RSP development checkout explicitly selects source mode through .rsp/source. Build the current source with mise exec -- pnpm run build, then use mise exec -- node dist/cli.mjs <command> from this checkout for every CLI operation. Skill examples use rsp as shorthand for this local invocation. Do not substitute a registry package, global CLI or another checkout, even when its version matches. Package version metadata is not source identity; retain the commit, uncommitted diff and build inputs for candidate evidence. Authorized update reconciles this checkout's managed environment from its local source and keeps source mode; it does not select a published version.

<!-- rsp:end -->

Project Development

  • mise.toml declares the maintainer Node, pnpm, Python and uv version ranges. Run repository tools through mise exec --; uv uses the mise Python without downloading a second interpreter. Dependency versions remain in pnpm-lock.yaml and tests/skills/uv.lock. The evaluation image builder resolves installed runtime versions through mise; Docker/Compose are host prerequisites.
  • Bundled assets in the repository root are the authored package sources.
  • This maintainer checkout uses mise exec -- node dist/cli.mjs <command> after building the source candidate. Do not replace it with the registry package merely because its version string matches; source behavior may be unreleased. Rules are retired; edit authored Skills, not discovery projections.
  • Required implementation coverage uses mise exec -- pnpm run build, mise exec -- pnpm run lint, and mise exec -- pnpm run test. Reuse traceable passing results while their relevant inputs and tested boundary remain unchanged; run only missing or invalidated coverage. For document/Skill-only changes, use affected package, resource, document and offline case checks rather than repeat unaffected code checks. Explicit execution or independent-acceptance requirements still apply.
  • Keep observable CLI, packaging and maintenance-tool regression tests under tests/code/; evaluate Skill behavior with native Harbor 0.24.0 tasks/jobs and harbor-rewardkit 0.2.3 under tests/skills/. Install the locked environment with uv sync --project tests/skills --frozen; run offline native task schema checks with uv run --project tests/skills python tests/skills/evaluate.py check. Package release aggregates local checks and only this offline Skill check; never start paid model calls automatically. Native jobs/default.yaml owns agent skills and job settings; follow tests/skills/README.md for separately authorized execution, calibration, resume and regrade. Keep private run evidence under ignored tests/skills/reports/ and selected research conclusions in time-named Markdown under research/reports/. Never reinterpret historical verdicts as current acceptance.
  • Keep tests focused on observable behavior and public command output rather than duplicated generated content or Skill prose fragments.

Project Skill Dogfooding

  • Use the repository-discovered rsp, rsp-shape, rsp-implement, rsp-doc, rsp-verify, rsp-review, and rsp-commit Skills for matching work in this checkout. Shape owns bounded read-only design; Implement owns diagnosis, justified test-first work, and authorized fixed-finding corrections. Review owns explicitly bounded report-only structural discovery through its conditional method. Core selects conditional coordination only for a real obligation, not ordinary single-owner continuity. This self-host uses workflow.delegation: off and workflow.closeout: ask. Explicitly required independent review remains a separate obligation. Ordinary completion and checkpoint commits keep work open; tracked commits with completed Tasks select the Skill's closeout decision even with verification gaps. Archive still requires authority and fresh gates. Configuration grants no product, Git, remote, publication, approval or human-acceptance authority.
  • These entries are live projections of the authored package sources under skills/; edit the authored source, not .agents/skills/.
  • Default rsp-doc owns repository-document, Skill and release writing through conditional artifact methods. rsp-review remains read-only, including release-document checks; both apply the writing-quality contract with standalone guidance.
  • Maintainer author-rsp-skills adds local Change, provenance and evaluation requirements to Doc's Skill method; these requirements do not apply to ordinary project Skills.
  • Treat host metadata limits as hard constraints, but do not use Skill body or reference word-count ceilings as correctness gates. Review concision semantically: remove duplication and unnecessary prose without losing trigger, authority, action, stop, return, or conditional-loading behavior.
  • Keep overlapping global engineering workflow Skills disabled in the maintainer environment while dogfooding so RSP capability gaps remain visible. Codex currently supports Skill disablement in user config, not project .codex/config.toml; do not add a misleading project-local disable list.

Maintainer Research

For upstream preparation, source distillation, or cross-source model synthesis, load the repo-local distill-upstream skill. Keep research under research/; promote selected recommendations through a normal RSP change.

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.