feat(skills): add bmad-os-findings-triage HITL triage skill

[alexeyv]

Summary

• Adds bmad-os-findings-triage skill: team-based HITL triage orchestrator for review findings
• One Opus agent per finding — researches autonomously, proposes plan, holds for human conversation
• Team lead orchestrates lifecycle: parsing, task creation, agent spawning, scorecard, cleanup
• Handles edge cases: wave batching (>25 findings), same-file conflicts, serialization, skip/defer

Files

• .claude/skills/bmad-os-findings-triage/SKILL.md — registration (7 lines)
• .claude/skills/bmad-os-findings-triage/prompts/instructions.md — team lead orchestration (287 lines)
• .claude/skills/bmad-os-findings-triage/prompts/agent-prompt.md — per-finding agent contract (105 lines)

Test plan

• Invoke skill with a structured findings report file
• Verify agents spawn in parallel and signal ready
• Verify HITL conversation flows naturally per finding
• Verify decisions are recorded and scorecard is accurate
• Test skip/defer flow
• Test with >25 findings to verify wave batching prompt

🤖 Generated with Claude Code

[github-actions]

@coderabbitai review

[augmentcode]

🤖 Augment PR Summary

Summary: This PR introduces a new Claude skill, bmad-os-findings-triage, to orchestrate human-in-the-loop (HITL) triage of review findings using a team-lead + per-finding agent pattern.

Changes:

• Adds skill registration metadata in .claude/skills/bmad-os-findings-triage/SKILL.md
• Adds a team-lead orchestration prompt that covers input parsing, task creation, team setup, agent spawning, readiness tracking, decision recording, batching for large finding sets, and cleanup
• Adds a per-finding agent prompt defining a strict lifecycle: autonomous research → visible plan → wait for HITL conversation → report a single final decision
• Defines a standardized decision taxonomy (FIX/WONTFIX/DISMISS/REJECT/SKIP/DEFER) and a scorecard format for final reporting

Technical Notes: The design emphasizes parallelism (one Opus agent per finding), explicit messaging conventions for readiness/decisions, and safeguards for >25 findings and same-file edit conflicts.

_{🤖 Was this summary useful? React with 👍 or 👎}

[augmentcode]

Review completed. 2 suggestions posted.

Comment augment review to trigger a new review at any time.

[alexeyv]

@CodeRabbit review

[coderabbitai]

✅ Actions performed

Review triggered.

Note: CodeRabbit is an incremental review system and does not re-review already reviewed commits. This command is applicable only when automatic reviews are paused.