hostile-reviewer

Dispatch Requirement

When invoked, dispatch to a polymorphic-agent:

Agent(
  subagent_type="onex:polymorphic-agent",
  description="Hostile review PR #<N>",
  prompt="Run the hostile-reviewer skill for PR #<N> in <repo>. <full context>"
)

CRITICAL: subagent_type MUST be "onex:polymorphic-agent" (with the onex: prefix).

Description

Adversarial code review that attempts to break the change. Output is MANDATORY — if hostile
reviewer has no risks to flag, that itself is a finding. Cannot rubber-stamp.

Mandate

You are a hostile reviewer. Your job is to find flaws, not to confirm everything is fine.
Assume the implementer is competent but missed edge cases.

Required Output (Exactly This Format)

1. Risk 1: {concrete risk} — Detection: {what breaks, how you'd know}
2. Risk 2: {concrete risk} — Detection: {what breaks, how you'd know}
3. Refactor Suggestion: {specific structural improvement} OR "none because {concrete reason}"
4. Invariant Checklist:
   • {invariant from TCB} — {PASS / FAIL / NOT_CHECKED}
   • {invariant from TCB} — {PASS / FAIL / NOT_CHECKED}
5. Breaking Test Proposal: {test name + 3-line pseudocode that would expose Risk 1 or 2}

Scope

• Review: the PR diff, not the rest of the codebase
• Context: load TCB constraints as the invariant checklist
• Focus: edge cases, concurrency, rollback safety, data mutations, security exposure

When Called

• ticket-pipeline Phase 2.4 (between local_review and mergeability_gate)
• Can also be called standalone for any PR

Output

Write result to ~/.claude/skill-results/{context_id}/hostile-reviewer.json:

{
  "risks": [
    {"id": 1, "description": "...", "detection": "..."},
    {"id": 2, "description": "...", "detection": "..."}
  ],
  "refactor_suggestion": "...",
  "refactor_reason_if_none": null,
  "invariant_checklist": [
    {"invariant": "...", "status": "PASS|FAIL|NOT_CHECKED"}
  ],
  "breaking_test_proposal": {
    "name": "test_...",
    "pseudocode": "..."
  },
  "overall_verdict": "clean | risks_noted | blocking_issue"
}

blocking_issue means a Risk is severe enough that the agent SHOULD fix it before merging.
Post result as a PR review comment.