Fixes #2

Transform pr-comment-responder from a monolithic workflow to a thin
coordination layer that delegates to specialized agents, following
the orchestrator pattern.

Key changes:
- Add Phase 2 Comment Triage with agent routing table
- Add Phase 3 Agent Delegation with example prompts for:
  - analyst: technical merit evaluation
  - critic: assumption validation
  - high-level-advisor: strategic decisions
  - implementer: code fixes
  - qa: regression verification
- Add Phase 4 Response Synthesis based on agent feedback
- Move bot-specific handling to Phase 5 with memory retrieval
- Add Agent Workflow table to output format
- Add Handoff Protocol table matching orchestrator style
- Remove embedded evaluation logic (delegate to agents)
- Maintain GitHub API interaction knowledge
- Update all three platforms (Claude, Copilot CLI, VS Code)

Benefits:
- Reuses existing agent expertise (analyst, critic, etc.)
- Reduces duplication of evaluation logic
- Follows orchestrator workflow pattern
- Token-efficient via delegation rather than embedded logic
- Easier to maintain and extend

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Fixes #2

Transform pr-comment-responder from duplicating workflow logic to a thin
triage layer that delegates to orchestrator with path recommendations.

Key changes:
- Remove duplicated workflow execution (orchestrator handles this)
- Add clear triage decision tree matching standard paths:
  - Quick Fix: implementer → qa (token-efficient)
  - Standard: analyst → architect → planner → critic → implementer → qa
  - Strategic: independent-thinker → high-level-advisor → task-generator
- Quick fixes go directly to implementer (skip orchestrator overhead)
- Standard/Strategic paths delegate to orchestrator with:
  - Classification and rationale
  - Code context and initial assessment
  - Workflow path recommendation
- Retain bot-specific handling (Copilot follow-up, CodeRabbit commands)
- Consistent across all three platforms

Triage signals:
- Quick Fix: "Can explain fix in one sentence"
- Standard: "Need to investigate first"
- Strategic: "Question is whether, not how"

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Apply lessons from Weinberg, Schrag, and Coding Horror articles to optimize
orchestrator resource usage across all three platforms (claude, copilot-cli,
vs-code-agents).

Key improvements:
- Add Phase 0: Triage with exit early criteria before orchestrating
- Add Complexity Assessment (Trivial/Simple/Standard/Complex) to right-size
  agent chains
- Add Value Checkpoint with Schrag's Principle after Phase 2
- Add Weinberg's Law of the Hammer quote to prevent over-orchestration
- Add Pattern Applied section to output format (Weinberg's Consulting Secret)
- Fix malformed markdown code fences in vs-code-agents version

Philosophy: "The cheapest orchestration is the one that doesn't happen."
Not every task needs every agent.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

… clarity

Address underutilization of key agents and strengthen memory as a triage asset.

## pr-comment-responder enhancements

### Memory as a Strength
- Add comprehensive memory protocol for bot false positive patterns
- Store reviewer preferences (bots AND humans have patterns)
- Track triage decisions (comment → path → outcome) for accuracy improvement
- Store successful rebuttals for confidence in declining suggestions
- Query domain-specific patterns for faster routing

### Domain-Specific Routing (skip orchestrator when appropriate)
- `.github/workflows/*` → devops (CI/CD domain expertise)
- `.githooks/*` → devops + security (infrastructure + security)
- `**/Auth/**`, `*.env*` → security (critical path)
- "WHETHER to do X" questions → independent-thinker (challenge assumptions)

### Independent-thinker Utilization
- Route "WHETHER" questions to independent-thinker BEFORE strategic path
- Provides unfiltered analysis of reviewer concerns
- Intellectually honest evaluation of tradeoffs

## Orchestrator enhancements

### Planner vs Task-Generator Clarity
- planner: Epic/Feature → Milestones (high-level breakdown)
- task-generator: PRD/Milestone → Atomic tasks (implementation-ready)
- Workflow: roadmap → planner → task-generator → implementer/qa/devops

### Routing Heuristics Updates
- "Challenge assumptions" now routes to independent-thinker (not critic)
- "Plan validation" explicitly routes to critic
- Clear distinction between Epic→Milestones and Milestones→Tasks

All three platforms (claude, copilot-cli, vs-code-agents) synced.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>