Review Packet

References

Read these first:

• ../../../.agents/references/context-layering-policy.md
• ../../../.agents/references/contract-gating-rule.md
• ../../../.agents/references/evidence-chain-rule.md
• ../../../.agents/references/lesson-quality-rule.md
• ../../../.agents/references/language-policy.md

Schemas:

• ../../../schemas/evidence_chain.schema.json
• ../../../schemas/review_packet.schema.json

When To Use

Use this skill when a Campaign reaches a review boundary — review cadence, Claim Delta, stop/abort decision, or budget exhaustion — and the write Operation must hand the operator a decision-ready artifact.

Required Work

1. Collect the Iteration's Light Conclusion Evidence, Research Briefs, semantic commits, metrics, and Gate refs.
2. Author either an Evidence Chain under .evidence/chains/ (validating against schemas/evidence_chain.schema.json) or a Review Packet under .evidence/review_packets/ (validating against schemas/review_packet.schema.json). The Campaign controller validates the artifact before entering waiting_review; a file that fails schema validation does not count.
3. Summarize only:
   • requested decision
   • verified facts vs inferences vs open questions
   • blocking items
   • exact human action needed
4. Stop at the human boundary. Resume, complete, or revise only after the operator's explicit decision.

Output Rules

• Do not mark any contract as accepted from the packet alone; approval is recorded only through harness campaign approve on explicit human instruction.
• Do not treat packet generation alone as approval or readiness.
• If the review surfaces blockers, propose the smallest next fix instead of expanding scope.