Repository Audit (Repo Explorer)

When to use

Run on every request before planning. Output is chat-only (no repo files). Keep it compact.

Inputs

• Working tree contents (paths + key files)
• Recent changes (if available)
• CI status signals (workflows present; failures if visible)
• /docs contents and relevance to current code

Output (exact format)

Return a Repo Survey with these sections (bullets, short lines):

1. Stack & Tooling
2. Docs State
3. Drift Checks (Docs↔Code, Skills↔Reality)
4. Quality Infrastructure
5. Suggested Tasks (titles + gates)

Step 1 — Identify stack & tooling

Look for (examples, not exhaustive):

• Languages: .csproj, package.json, pyproject.toml, requirements.txt, go.mod, pubspec.yaml
• Build: Makefile, docker-compose.yml, *.sln, gradle, mvnw, nx.json, turbo.json Report:
• Primary stack(s)
• Build entrypoints (Make targets preferred)
• Package manager(s)

Step 2 — Check docs presence & shape

Inspect /docs:

• Does it exist?
• Is there /docs/use-cases/? Are use cases one-per-file with UC-NNN naming?
• Is there /docs/architecture/? Are decisions documented with ARCH-NNN naming?
• Is there /docs/testing/? Are test strategies documented with TEST-NNN naming? Report:
• What's present
• What's missing (templates exist at *-000-template.md in each folder)
• Any obvious staleness/duplication

Step 3 — Drift checks (blocking signals)

Docs ↔ Code drift (blocking)

Look for mismatches such as:

• Documented endpoints that don’t exist
• Implemented endpoints/flows not documented
• Behavior described differently than code/tests imply Return: drift_docs_code: none | suspected | confirmed with 1–2 bullet reasons.

Skills ↔ Reality drift (blocking)

Compare repo reality against skills assumptions:

• Lint tool differs from skill
• Test runner differs
• CI structure differs
• Docker/Make conventions differ Return: drift_skills: none | suspected | confirmed with 1–2 bullet reasons.

Step 4 — Quality infrastructure inventory

CI (blocking if missing/failing)

• Check .github/workflows/
• Identify primary workflow(s): lint/build/test
• If evidence of failing CI exists, report it Return:
• ci: missing | present | failing

Lint/format

• Identify lint tools and where configured Return:
• lint: missing | present and the tool

Tests

• Identify unit/integration/e2e presence
• Note Playwright presence for web Return:
• tests: none | unit | unit+integration | unit+e2e | full (best match)

Dockerization (required for HTTP APIs)

• Check for docker-compose.yml and .env.example Return:
• docker: missing | present and whether it includes the API

Makefile (preferred local UX)

• Check for Makefile
• List key targets if present: up/down/test/lint/build Return:
• make: missing | present and key targets

Step 5 — Suggested Tasks (orchestrator will create real tasks)

Provide 5–12 suggested tasks max, each as:

• Title
• Why (one short line)
• Suggested required_gates[] (explicit list)

Rules:

• If drift_docs_code != none: include “Resolve docs↔code drift” as the top task (blocking).
• If ci != present: include “Add/Fix CI” as a top task (blocking).
• If HTTP API and docker: missing: include “Add docker-compose stack” (blocking for use-case work).
• If drift_skills != none: include “Update affected skills to match reality” (blocking).

References (read/consult)

• AGENTS.md (non-negotiables)
• .github/copilot-instructions.md (repo instructions)
• .github/skills/gaia-process/SKILL.md (workflow contract)
• .github/workflows/ (CI)
• /docs/ (source of truth)
• Makefile (local commands)