docs: define evidence-gated Hermes handoff loop for Claw Code execution

[Yeachan-Heo]

Source

Discord #claw-code SDLC loop proposal: Hermes → handoff packet → Claw Code/subagents → evidence bundle → Hermes post-flight → commit/test.

Problem

Agentic SDLC loops fail when each stage reinterprets the task. The handoff packet should be treated as a contract boundary, not a nice summary.

Suggested docs/spec

Document a durable handoff loop:

1. Hermes intake
2. Handoff packet with explicit scope/success criteria
3. Claw Code reads contract / blueprint / .omx / CURRENT_STATE
4. Execution mode selection:
   • simple
   • $ralplan
   • $ultrawork
   • $ralph
   • $team
5. Narrow lane execution
6. Evidence bundle:
   • diff
   • changed-files-summary
   • test logs
   • verification outputs
   • subagent reports
   • execution report
7. Hermes post-flight checks evidence, not vibes
8. Gate result:
   • PASS
   • KNOWN_RISK_ACCEPTED
   • BLOCKED
   • FAIL
9. Commit only if gate permits

Acceptance criteria

• A handoff packet template exists.
• Required evidence artifacts are listed.
• Commit/merge gates depend on evidence, not agent confidence.

—
[repo owner's gaebal-gajae (clawdbot) 🦞]