AGENTS.md

AGENTS.md

Codex + review agents. eRPC repo (Go + TS + infra/config).

Autonomous Harness Protocol

• Start: read this file + review/README.md.
• Build context: scripts/agent-harness/context.sh (--staged or <range> optional).
• Pick runbook:
  • review/runbooks/bugfix.md
  • review/runbooks/feature.md
  • review/runbooks/review.md
• Keep docs current when code/config/contracts shift. Use review/autonomy.md.
• Before handoff:
  • make agent-check
  • make agent-review-load
  • make agent-gate (or make agent-gate-full)

Harness Artifacts

• Entry + guardrails: AGENTS.md
• Harness index: review/README.md
• Skills+shell guidance: review/skills-shell-tips.md
• Repo map: review/repo-map.md (generated)
• Contract surfaces: review/contracts.md
• Stateful/data invariants: review/data-models.md
• Severity rubric: review/checklist.md
• Maintenance triggers: review/autonomy.md
• Mechanical scripts: scripts/agent-harness/*.sh
• Artifact boundary: artifacts/agent/
• Scheduled loops: .github/workflows/agent-automations.yml

Review Depth Protocol (Repo-Aware)

• Read review/README.md + review/checklist.md first.
• Run scripts/review-repo-map.sh for current repo map.
• Run scripts/review-impact-map.sh <range> or --staged for impact map.
• Findings: issue index by severity + confidence scores + low-confidence watchlist.

Severity

• CRITICAL: Security, data loss, breaking API behavior, auth bypass.
• HIGH: Reliability, correctness, consensus integrity, cache corruption.
• MEDIUM: Performance, observability gaps, config drift.
• LOW: Style/docs nits.

Review Focus (eRPC-specific)

• JSON-RPC correctness + error normalization.
• Re-org aware cache correctness + invalidation.
• Consensus policy + integrity checks.
• Upstream selection, retries, timeouts, hedging, circuit breakers.
• Rate limiting + cost control.
• Auth flows (basic/JWT/SIWE), secret handling.
• Telemetry/metrics/logging (Prometheus, tracing).
• Config schema changes (erpc.yaml, erpc.dist.*).

Canonical Commands

• Context: make agent-context
• Refresh map: make agent-refresh
• Harness checks: make agent-check
• Direct check script: scripts/agent-harness/check.sh
• Skills+shell check: scripts/agent-harness/skills-shell-check.sh
• Review bottleneck budget: make agent-review-load
• Direct review-load script: scripts/agent-harness/review-load.sh
• PR mergeability scan: make agent-pr-health
• Random latent-bug sweep: make agent-random-bug
• Contribution digest: make agent-digest
• Artifact boundary path: artifacts/agent/
• Full autonomous gate:
  • make agent-gate (build + fast tests)
  • make agent-gate-full (build + full tests)