agentleFS
emiliodominguezemiliodominguez/flow-skills/skills/flow-repo-brief/SKILL.md

flow-repo-brief

Skill1 starsChanged 50 days agoMITView on GitHub

Maintained, source-cited brief for a subsystem that agents reuse across sessions. Use when agents keep rediscovering conventions and commands, or agent notes on a subsystem need refreshing against source. Feeds `flow-plan` and `flow-specialize`.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 2 are distinctive, and 1 is 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.

  • 1

    Very specific

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

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

    Write to the user's knowledge location, else .plans/context/<topic>.md.

    line 19
  2. Distinctive:

    Keep it small enough to use directly.

    line 42
  3. Distinctive:

    Add scope, architecture, commands, conventions, hypotheses and refresh-trigger sections only when task-relevant.

    line 48

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

Repository brief

Capture the facts that change implementation decisions, with sources another agent can reopen. A brief is maintained evidence, never an authority above user or repository instructions.

Phase 1: Bound the question

  • Identify the task or subsystem and the next consumer. Read agent instruction files (for example AGENTS.md), manifests, lockfiles, lint/format config, CI and representative nearby code. Trace one real flow before describing architecture. Use flow-onboard first if a broad tour is needed; do not scan the whole repository for a narrow change.
  • Write to the user's knowledge location, else .plans/context/<topic>.md. Preserve tracked or ignored status and say when an ignored file is local-only. Do not change global instructions, ignore rules or installed skills as a side effect.

Phase 2: Separate fact from inference

Cite a path plus symbol or line range at the recorded revision for every load-bearing claim; add command output for operational claims. Capture:

  • Repository identity, branch, commit, date, scope, relevant dirty/untracked files.
  • Constraint provenance: explicit instruction, executable config, repeated precedent, or inference. One example is not a universal rule.
  • Entry points, ownership boundaries, key types/contracts, one real data flow.
  • Exact build/test/lint commands and cwd, each marked run, observed in config, or unavailable, with actual outcomes.
  • Reusable patterns, exceptions, failure modes and areas the task must not change.
  • Assumptions and hypotheses, each with the evidence that would confirm or falsify it.

Record variable names and required access, never secrets or private payloads. For uncertain external API facts, prefer primary docs or pinned source and record the version and date.

Phase 3: Write for selective reading

Keep it small enough to use directly. Use a claims table:

ClaimKindSource at revisionConfidence / unresolved evidence
Function declarations and four-column tabsconfig or instructionactual file and keyconfirmed or unknown

Add scope, architecture, commands, conventions, hypotheses and refresh-trigger sections only when task-relevant. Split large subsystem detail into linked sibling files with an index of what to read per task. No copied source, transcripts, file listings or framework tutorials.

Phase 4: Refresh and verify

Reopen the sources behind claims the next task relies on. Compare current instructions, config, source and dependency versions with recorded inputs: changed cited inputs invalidate claims, an unrelated commit does not. Mark unsupported claims stale and update only affected sections. A date alone never proves old evidence valid. For consequential decisions, have an independent pragmatic verifier check claims and hypotheses via flow-verify; missing verification stays visible.

Anti-patterns

  • "Uses DDD" or "all IDs are opaque" without types and boundary evidence.
  • Copying another repository's preferences over this one's conventions.
  • Calling a command verified when it was only found in package.json.
  • Treating a stale brief as permission to ignore live source or user scope.

Done when

  • The next worker can find relevant code, conventions, commands and open questions.
  • Load-bearing claims have citations and a current or explicitly stale status.
  • The artifact path and refresh conditions are recorded.

Hand off to flow-plan for decomposition or flow-specialize for a repository-specific worker.

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.