agent-readiness（Agent 准备情况）

Agent-Readiness

Make a repo ready for autonomous agent work.

Principles

• Environment > instruction — infrastructure matters more than the prompt
• Mechanical enforcement > prose — hooks, CI, health checks, and scripts beat wishes
• Separate builder from judge — agent-readiness builds the rig, verify proves your own change, review critiques existing code
• Real behavior > mocked confidence — smoke, integration, and e2e checks beat large suites that mostly mock the seam under test
• Smallest useful layer first — add layers in order, stop when the repo becomes reliably verifiable
• Progressive disclosure — keep the core workflow here, load patterns on demand

Handoffs

• Need to review existing code, a diff, branch, or PR → use review
• Need to prove your own completed change works on real surfaces → use verify
• Need to update AGENTS.md, README.md, specs, or repo docs → use docs

The 7-Layer Stack

1. Boot — single command starts the app
2. Smoke — a fast proof the app is alive
3. Interact — agent can exercise the real surface
4. E2e — key user flows work end to end
5. Enforce — hooks, CI gates, lint rules, or mechanical checks
6. Observe — logs, health endpoints, traces, machine-readable signals
7. Isolate — worktrees or containers do not collide

Concrete examples:

• Boot: pnpm dev, cargo run, or docker compose up
• Smoke: curl http://127.0.0.1:3000/health
• Interact/E2e: pnpm exec playwright test
• Observe: structured logs or a machine-readable health endpoint

Workflow

1. Audit

Grade the repo across these dimensions:

• bootable
• testable
• observable
• verifiable

For each, report:

• status: pass / partial / fail
• evidence: file or command
• gap: what is missing

Use references/grading.md. Lowest dimension sets the overall grade.

Example output:

bootable: partial — `pnpm dev` starts the app after manual env setup
testable: fail — only mocked tests under test/
observable: partial — health endpoint exists, structured logs missing
verifiable: fail — no stable smoke or interaction script
overall grade: D

2. Setup

Build missing layers in this order:

Boot → Smoke → Interact → E2e → Enforce → Observe → Isolate

Each step should be independently useful. Stop once the repo is reliably verifiable; do not build a cathedral because you got excited.

When readiness work includes agent entrypoints, keep AGENTS.md as the canonical authored guide and place CLAUDE.md beside it as a symlink to AGENTS.md rather than maintaining two separate guidance files.

Boot — create a single-command entry point:

#!/usr/bin/env bash
set -euo pipefail
<your-boot-command> &
APP_PID=$!
for i in $(seq 1 30); do
  curl -sf http://localhost:${PORT:-3000}/health > /dev/null 2>&1 && break
  sleep 1
done
curl -sf http://localhost:${PORT:-3000}/health > /dev/null 2>&1 || {
  echo "ERROR: App failed to start"; kill $APP_PID 2>/dev/null; exit 1
}
echo "App is ready"

Smoke — fast proof the app is alive (< 5 seconds):

curl -sf http://localhost:3000/health | jq .   # HTTP service
./dist/my-cli --version                         # CLI tool
npx playwright test smoke.spec.ts               # UI app

Enforce — pre-push hook to catch failures before CI:

#!/usr/bin/env bash
# .git-hooks/pre-push
set -euo pipefail
<your-lint-command>
<your-smoke-command>

See references/setup-patterns.md for e2e, observability, isolation, and containerized stack patterns.

3. Improve

Tighten weak or flaky layers:

• remove mock-only confidence theater
• prefer smoke, integration, and e2e checks over mock-heavy suites that self-verify implementation details
• replace one-off checks with reusable scripts or hooks
• add dead-code or unused-symbol enforcement where the stack supports it
• add logs and health signals agents can query
• make parallel work safe when agent collisions are real

4. Hand Off

When the repo reaches C+ and can be judged honestly, hand off to verify or review. If changes created doc drift, hand off to docs.

Anti-Patterns

• Mock-only tests — pass by construction, verify nothing
• Mock-heavy unit suites as the main proof — agents love them because they are easy to satisfy, not because they prove the system works
• Self-evaluation — builder grading its own work
• Docs-only fixes disguised as readiness work
• Routine PR review here — that's review
• Perfect infrastructure upfront — iterate from real failure modes

Output

After readiness work, report:

• grade before and after
• dimensions with evidence
• files changed
• remaining gaps ranked by impact
• verify readiness
• recommended next handoff: verify, review, docs, or human review

References

• references/grading.md — agent-readiness grading scale with mechanical criteria
• references/setup-patterns.md — boot, smoke, e2e, observability, and isolation patterns
• references/industry-examples.md — external patterns and justification for readiness investment