AGENTS.md

Agent Instructions (Source of Truth)

This file is the canonical instruction entrypoint for local agent tooling in this repo.

Read Order

1. This file (high-signal defaults and routing)
2. .claude/guidelines/beads-rules.md
3. .claude/programming-guidelines.md
4. .claude/contracts
5. .claude/skills
6. .claude/test-scenarios/README.md
7. .github/PULL_REQUEST_TEMPLATE.md

Must-Follow Defaults

• Use develop as the base branch.
• Use pnpm.
• Keep changes surgical and focused.
• For complex multi-step tasks, prefer bd tracking before implementation.
• Prefer origin for new feature/fix branch pushes when permissions allow.
• If a PR branch is already on fork, keep using that existing fork branch.
• Run pnpm run lint --fix and pnpm run type-check after code changes.
• Use .github/PULL_REQUEST_TEMPLATE.md for PR bodies.

Routing

• New second-class EVM chain work: .claude/contracts/second-class-evm-chain.md
• New swapper integration work: .claude/contracts/swapper-integration.md
• Test scenarios (/test-agent, @neomaking flow): .claude/test-scenarios/README.md
• QA automation (qabot, fixture-based flow): .claude/skills/qabot/SKILL.md
• Translation workflow: .claude/skills/translate/SKILL.md

Tooling Notes

• Claude compatibility is provided by CLAUDE.md importing this file.
• Codex compatibility is pinned via .codex/config.toml.
• Full legacy CLAUDE.md policy content was moved to .claude/programming-guidelines.md to avoid loss while keeping top-level entrypoints compact.
• Repo-local skills live in .claude/skills.
• Additional user-level skills/config may exist in ~/.agents, ~/.codex/skills, and ~/.openclaw.