feat(agents): add skill-pr-addresser agent for PR feedback processing

.claude/agents/skill-pr-addresser.md

@@ -0,0 +1,217 @@
+---
+description: Address PR review feedback for skills, continuing development until approved
+tools: [Read, Write, Edit, Bash, Glob, Grep, Task]
+---
+
+# Skill PR Addresser
+
+Address PR review feedback for skills, continuing development until the PR is approved.
+
+## Overview
+
+This agent picks up where `skill-reviewer` leaves off. When a PR has review feedback requesting changes, this agent:
+
+1. **Discovers** PRs with pending feedback (blocking reviews, unresolved threads)
+2. **Analyzes** the feedback using the `feedback-analyzer` sub-agent
+3. **Fixes** issues using the `feedback-fixer` sub-agent with model escalation
+4. **Commits** and pushes changes to the PR branch
+5. **Comments** on the PR summarizing what was addressed
+6. **Requests re-review** when all feedback is addressed
+7. **Repeats** until approved or max iterations reached
+
+## Usage
+
+### Address feedback on a specific PR
+
+```bash
+just -f .claude/agents/skill-pr-addresser/justfile address 795
+```
+
+### Check status of a PR
+
+```bash
+just -f .claude/agents/skill-pr-addresser/justfile status 795
+```
+
+### With options
+
+```bash
+# Dry run - show what would be done without making changes
+just -f .claude/agents/skill-pr-addresser/justfile address-dry 795
+
+# Specify skill explicitly (instead of auto-detecting from changed files)
+just -f .claude/agents/skill-pr-addresser/justfile address-skill 795 components/skills/lang-rust-dev
+
+# Force addressing even if no pending feedback detected
+just -f .claude/agents/skill-pr-addresser/justfile address 795 --force
+```
+
+### Session management
+
+```bash
+# List active sessions
+just -f .claude/agents/skill-pr-addresser/justfile sessions
+
+# List all sessions (including completed)
+just -f .claude/agents/skill-pr-addresser/justfile sessions-all
+```
+
+## Sub-Agents
+
+| Agent               | Model      | Purpose                              |
+| ------------------- | ---------- | ------------------------------------ |
+| `feedback-analyzer` | Haiku 3.5  | Parse and categorize feedback items  |
+| `feedback-fixer`    | Sonnet 4   | Implement fixes in skill files       |
+
+### Model Escalation
+
+The fixer uses automatic model escalation:
+- **Simple nitpicks** (2 or fewer): Haiku 3.5
+- **Complex changes**: Sonnet 4
+
+## Architecture
+
+```
+skill-pr-addresser/
+├── skill-pr-addresser.md      # This file (discovered by Claude Code)
+├── main.py                    # CLI entry point
+├── justfile                   # Task runner recipes
+├── pyproject.toml             # Python package config
+├── src/
+│   ├── app.py                 # Cement CLI application
+│   ├── discovery.py           # PR and session discovery
+│   ├── github_pr.py           # GitHub PR operations
+│   ├── feedback.py            # Feedback analysis/fixing
+│   ├── addresser.py           # Main orchestration loop
+│   ├── templates.py           # Mustache template rendering
+│   └── exceptions.py          # Custom exceptions
+├── subagents/
+│   ├── feedback-analyzer/     # Analyzes feedback into items
+│   │   ├── prompt.md
+│   │   └── config.yml
+│   └── feedback-fixer/        # Implements fixes
+│       ├── prompt.md
+│       └── config.yml
+├── templates/
+│   ├── iteration_comment.hbs  # PR comment for each iteration
+│   ├── ready_comment.hbs      # "Ready for review" comment
+│   └── skipped_feedback.hbs   # Skipped items summary
+├── data/
+│   ├── config.json            # Runtime configuration
+│   └── sessions/              # Session state (gitignored)
+└── tests/
+    └── *.py                   # Unit tests
+```
+
+### Shared Library
+
+Uses `skill-agents-common` for:
+- GitHub operations (`github_ops.py`)
+- Worktree management (`worktree.py`)
+- Session tracking (`session.py`, `models.py`)
+
+### Worktree Reuse
+
+Reuses worktrees created by `skill-reviewer` when available:
+- Pattern: `/private/tmp/worktrees/<project_id>/issue-<number>/`
+- Uses `get_or_create_worktree()` to find or create
+
+### Session Continuity
+
+Links to original `skill-reviewer` session:
+- Finds by issue number: `find_session_by_issue()`
+- Finds by PR number: `find_session_by_pr()`
+- Creates new if needed: `create_session_from_pr()`
+- Tracks iterations and results in session data
+
+## Feedback Loop
+
+```
+┌──────────────────────────────────────────────────────────────────┐
+│                    Feedback Addressing Loop                       │
+├──────────────────────────────────────────────────────────────────┤
+│                                                                   │
+│  ┌─────────────┐    ┌─────────────┐    ┌─────────────────────┐   │
+│  │  Discovery  │───▶│  Analyzer   │───▶│   Fixer (Haiku/    │   │
+│  │   (gh CLI)  │    │  (Haiku)    │    │    Sonnet)         │   │
+│  └─────────────┘    └─────────────┘    └─────────────────────┘   │
+│         │                                        │               │
+│         │                 ┌──────────────────────┘               │
+│         │                 │                                       │
+│         │                 ▼                                       │
+│         │         ┌─────────────────┐                            │
+│         │         │  Commit & Push  │                            │
+│         │         │    (git CLI)    │                            │
+│         │         └─────────────────┘                            │
+│         │                 │                                       │
+│         │                 ▼                                       │
+│         │         ┌─────────────────┐                            │
+│         │         │  Post Comment   │                            │
+│         │         │    (gh CLI)     │                            │
+│         │         └─────────────────┘                            │
+│         │                 │                                       │
+│         │                 ▼                                       │
+│         │         ┌─────────────────┐                            │
+│         └────────▶│ Check Complete? │◀───────────────────────────│
+│                   │  success >= 90% │                            │
+│                   └─────────────────┘                            │
+│                           │                                       │
+│                    Yes    │    No (iterate)                      │
+│                   ┌───────┴───────┐                              │
+│                   ▼               ▼                              │
+│          ┌────────────────┐    (loop)                            │
+│          │ Request Review │                                      │
+│          │    (gh CLI)    │                                      │
+│          └────────────────┘                                      │
+│                                                                   │
+└──────────────────────────────────────────────────────────────────┘
+```
+
+## Configuration
+
+### Environment Config
+
+See `config/skill-pr-addresser.conf`:
+
+```ini
+[skill-pr-addresser]
+repo_owner = aRustyDev
+repo_name = ai
+max_iterations = 3
+rate_limit_delay = 1.0
+worktree_base = /private/tmp/worktrees
+
+[otel]
+enabled = false
+endpoint = http://localhost:4317
+service_name = skill-pr-addresser
+```
+
+### Runtime Config
+
+Edit `data/config.json` for batch settings:
+
+```json
+{
+  "review_labels": ["review", "skills"],
+  "max_parallel": 3,
+  "max_cost_per_pr": 2.0
+}
+```
+
+## Cost Estimation
+
+Per PR addressing cycle (typical):
+
+| Model      | Calls | Cost/call | Total   |
+| ---------- | ----- | --------- | ------- |
+| Haiku 3.5  | 1-2   | ~$0.02    | ~$0.04  |
+| Sonnet 4   | 1-2   | ~$0.35    | ~$0.70  |
+| **Total**  |       |           | ~$0.75  |
+
+Complex PRs with multiple iterations: ~$1-2
+
+## Related
+
+- **skill-reviewer**: Creates the initial PR from issues
+- **skill-agents-common**: Shared library for both agents