phpunit-unit-test-debating

name: phpunit-unit-test-debating version: 3.7.2 description: Internal sub-skill. Do not auto-activate. Use only when explicitly invoked by name by another skill or agent. user-invocable: false allowed-tools: Read, Glob, Grep, SendMessage, mcp__plugin_test-writing_test-rules__get_rules

PHPUnit Test Review Debate

Peer-to-peer debate of review findings within a single Agent Teams wave. You debate directly with co-reviewers via SendMessage, then produce your final stance.

Input

Provided in spawn prompt by team-lead:

• own_findings: your findings from Wave 0 (reviewing skill output)
• peer_findings: findings from co-reviewers on shared files
• co_reviewers: list of co-reviewer names and shared files

Phase 1: Prepare Positions

Load references/debate-rules.md.

For each file, compare own findings against peer findings:

1. Shared findings (both you and peer reported) — prepare endorsement
2. Peer-only findings (peer reported, you didn't) — prepare challenge with detection algorithm citation, OR concession
3. Own-only findings (you reported, peer didn't) — prepare justification with code evidence

For challenges: call mcp__plugin_test-writing_test-rules__get_rules(ids={rule_id}) to load the detection algorithm. Apply it against the code. If the peer is right, prepare concession instead.

Phase 2: Debate (max 2 rounds)

Round 1

For each co-reviewer, send ONE message via SendMessage(to: "{co_reviewer_name}") covering all shared files. Use the debate message format from references/output-format.md:

• Endorsements for shared findings
• Challenges with detection algorithm citations
• Justifications for own-only findings with code evidence
• Concessions where peer evidence is stronger
• Cross-file references where applicable (first-hand only)

After sending, wait for co-reviewer responses.

Round 2 (if needed)

If you received challenges from co-reviewers in round 1:

1. Evaluate each challenge against the detection algorithm
2. Concede where their evidence is stronger
3. Defend where your evidence holds, citing specific code

Send ONE response per co-reviewer. Then proceed to Phase 3.

If no challenges were received, or all challenges are conceded, skip round 2.

Convergence

• After round 2: proceed to Phase 3
• After round 1 with no open challenges: proceed to Phase 3
• If co-reviewer does not respond: proceed to Phase 3 with available input
• Do NOT send more than 2 debate messages per co-reviewer

Phase 3: Final Stance

Produce final stance per file using the format from references/output-format.md:

• All findings you still stand by (with enforce level, location, current, suggested)
• Withdrawn findings with reasons (citing the peer's argument that convinced you)
• Cross-file references used during debate

This is your output. Return it to the lead.

Troubleshooting

Co-Reviewer Unresponsive

If a co-reviewer does not respond to your round 1 message, produce final stance from your own analysis and whatever peer findings you received in the input. Do not block.

MCP Tool Unavailability

If mcp__plugin_test-writing_test-rules__get_rules is unavailable, you cannot verify detection algorithms. Concede peer findings you cannot verify and note the limitation in your final stance.