agentleFS
ElguajoElguajo/Progressive-Context-Kit/AGENTS.md

Progressive-Context-Kit

AGENTS.md0 starsChanged 41 days agoMITView on GitHub

Universal engineering behavior is supplied by the user-global layer. Standalone composes it; router owns context, workflow/tool routing, project state. Load only matching Skills/protocols: - Routed .agents/... paths are repository-root-relative: project Skills win; missing is an error, no global fallback. Other global Skills work. - new product initialization → project-bootstrap - existing repository adoption → existing-project-adoption - missing materially useful preferred tooling → tooling-bootstrap - non-trivial implementation after direction is clear → implementation-execution - material architecture/technology fork → architecture-decision - auth/payments/permissions/secrets/private…

The file

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

3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    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:

    Normal product work: prefer python3 tools/context_compile.py.

    line 8
  2. Very specific:

    Never warm up with full completed phases, all ADRs, docs/system/*, docs/system/LINEAGE.md, full chat history, or large manuals.

    line 9
  3. Distinctive:

    If Git is unavailable, continue without treating that as an error.

    line 11

≈ “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

Progressive Context Kit — Personal Repository Router

Universal engineering behavior is supplied by the user-global layer. Standalone composes it; router owns context, workflow/tool routing, project state.

Context routing

  • Tiny/local task: target file + nearby evidence/tests only; project docs only when product/architecture constraints matter.
  • Normal product work: prefer python3 tools/context_compile.py; otherwise read docs/project/PROJECT_BRIEF.md → ARCHITECTURE.md → ROADMAP.md → NEXT_SESSION.md when present → [>] phase plan + active prompt (check freshness) → prior phase's Completion Record when present, then relevant ADR/source/tests/schemas/docs.
  • Never warm up with full completed phases, all ADRs, docs/system/*, docs/system/LINEAGE.md, full chat history, or large manuals.
  • ROADMAP.md is canonical for current phase. All [x] means new work is a change request.
  • If Git is unavailable, continue without treating that as an error.

Workflow routing

Load only matching Skills/protocols:

  • Routed .agents/... paths are repository-root-relative: project Skills win; missing is an error, no global fallback. Other global Skills work.
  • new product initialization → project-bootstrap
  • existing repository adoption → existing-project-adoption
  • missing materially useful preferred tooling → tooling-bootstrap
  • non-trivial implementation after direction is clear → implementation-execution
  • material architecture/technology fork → architecture-decision
  • auth/payments/permissions/secrets/private data/untrusted input/SQL/CSRF/redirects/webhooks/migrations/destructive work → security-sensitive-change
  • unclear/intermittent/stateful root cause → systematic-debugging
  • code/diff/PR review or pasted code without a specific question → code-review
  • material durable-governance documentation edit → documentation-governance
  • session ending or user-only decision/blocker → session-handoff
  • unclear/inconsistent project state → project-doctor

For implementation completion use docs/system/QUALITY_PROTOCOL.md. Installed Skills are not warm-up context.

Preferred tooling

Preferred implementations are explicit: Semble for intent/semantic discovery, Serena for known-symbol navigation/refactor, RTK for compact shell output, Superpowers for implementation/TDD/debug discipline, gstack for challenge/review/browser QA/release checks, Context7 for fresh library/API docs, and GitHub Spec Kit for optional Advanced Spec Mode. Read integrations/TOOL_REGISTRY.json / PROFILES.md only when selecting, checking, installing, or routing tools.

If a preferred tool is absent and materially useful, use tooling-bootstrap: verify current official installation docs, explain the benefit/permissions, and request one focused approval before installing or modifying user/global agent configuration. Do not interrupt a tiny task just to install tooling. Installed ≠ loaded ≠ invoked. One discovery question gets one primary route; a second tool must answer a different question, resolve ambiguity, or be fallback.

Canonical project state

Brief owns product outcome/scope; Architecture owns system shape/boundaries; Roadmap owns phase order/status; Phase owns tasks/acceptance/verification and prompt/evidence records; completed Phase owns its compact Completion Record; CONTEXT_MANIFEST.json owns optional phase hints; ADR owns consequential rationale; NEXT_SESSION owns hot navigation/prompt pointer; TOOLING_STATUS.json is tooling cache. Placement: docs/system/LAYER_OWNERSHIP.md.

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.