Skip to content

agents.md

Latest commit

 

History

History
62 lines (53 loc) · 2.42 KB

agents.md

File metadata and controls

62 lines (53 loc) · 2.42 KB
Multi-agent coordination for Starkclaw (agentic-native, fast iteration, frequent integration). | Role | Owns | Does NOT | | ---- | ---- | -------- | | Coordinator | milestone selection, task breakdown, `STATUS.md` accuracy, integration | implement large features | | Mobile Executor | Expo app UI/UX + storage + networking | change contract logic without coordination | | Contracts Executor | Cairo contracts + tests + deploy scripts | modify mobile UX without coordination | | Agent Runtime Executor | LLM adapter + tool runtime + audit log | change contract interfaces silently | | Reviewer | correctness, security rails, regression checks | do initial implementation | For any work item: 1. Investigation (Coordinator): identify files, interfaces, acceptance checks, risks, and a minimal plan. 2. Execution (Executor): implement exactly the plan, keep changes small, surface blockers immediately. 3. Review (Reviewer): run checks, scan for security regressions, confirm acceptance criteria.

<task_states> todo -> inprogress -> inreview -> done blocked can be entered from any state when waiting on a decision/dependency </task_states>

SAFE to parallelize: - `apps/mobile/**` work vs `contracts/**` work when interfaces are stable - UI iteration while contracts tests run (different files)

MUST serialize:

  • Any change that touches a shared interface (ABI, calldata shape, policy fields)
  • Changes to scripts/** and CI pipelines

Conflict resolution:

  1. Detect overlap early (rg for touched files).
  2. Pause later task if it touches the same interface.
  3. Land the interface change first with tests.
  4. Rebase/adjust dependents and re-verify.

<integration_protocol>

  • Default to short-lived branches only when risk is high; otherwise commit directly to main for speed.
  • Each milestone should land as a small series of commits aligned to acceptance criteria.
  • After each push: update STATUS.md so another agent can pick up instantly. </integration_protocol>
Escalate when: - A decision affects public interfaces (policy model, signing stack, package manager) - Security implications (key handling, injection/tool abuse, approval semantics) - You cannot make progress without user-provided info (RPC URL, funding, API keys)

Format:

Escalation: [Title]

Blocker: [...] Options:

  1. [...]
  2. [...] Recommendation: [...]