@paw-workflow/cli 0.0.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (38) hide show
  1. package/README.md +124 -0
  2. package/bin/paw.js +82 -0
  3. package/dist/agents/PAW-Review.agent.md +86 -0
  4. package/dist/agents/PAW.agent.md +171 -0
  5. package/dist/skills/paw-code-research/SKILL.md +209 -0
  6. package/dist/skills/paw-docs-guidance/SKILL.md +163 -0
  7. package/dist/skills/paw-git-operations/SKILL.md +196 -0
  8. package/dist/skills/paw-impl-review/SKILL.md +178 -0
  9. package/dist/skills/paw-implement/SKILL.md +153 -0
  10. package/dist/skills/paw-init/SKILL.md +118 -0
  11. package/dist/skills/paw-plan-review/SKILL.md +117 -0
  12. package/dist/skills/paw-planning/SKILL.md +217 -0
  13. package/dist/skills/paw-pr/SKILL.md +157 -0
  14. package/dist/skills/paw-review-baseline/SKILL.md +268 -0
  15. package/dist/skills/paw-review-correlation/SKILL.md +307 -0
  16. package/dist/skills/paw-review-critic/SKILL.md +373 -0
  17. package/dist/skills/paw-review-feedback/SKILL.md +437 -0
  18. package/dist/skills/paw-review-gap/SKILL.md +639 -0
  19. package/dist/skills/paw-review-github/SKILL.md +336 -0
  20. package/dist/skills/paw-review-impact/SKILL.md +569 -0
  21. package/dist/skills/paw-review-response/SKILL.md +118 -0
  22. package/dist/skills/paw-review-understanding/SKILL.md +372 -0
  23. package/dist/skills/paw-review-workflow/SKILL.md +239 -0
  24. package/dist/skills/paw-spec/SKILL.md +257 -0
  25. package/dist/skills/paw-spec-research/SKILL.md +138 -0
  26. package/dist/skills/paw-spec-review/SKILL.md +101 -0
  27. package/dist/skills/paw-status/SKILL.md +160 -0
  28. package/dist/skills/paw-transition/SKILL.md +134 -0
  29. package/dist/skills/paw-work-shaping/SKILL.md +99 -0
  30. package/dist/skills/paw-workflow/SKILL.md +142 -0
  31. package/lib/commands/install.js +103 -0
  32. package/lib/commands/list.js +18 -0
  33. package/lib/commands/uninstall.js +95 -0
  34. package/lib/commands/upgrade.js +119 -0
  35. package/lib/manifest.js +42 -0
  36. package/lib/paths.js +42 -0
  37. package/lib/registry.js +41 -0
  38. package/package.json +40 -0
@@ -0,0 +1,163 @@
1
+ ---
2
+ name: paw-docs-guidance
3
+ description: Documentation conventions for PAW implementation workflow. Provides Docs.md template structure, include/exclude guidelines, and project documentation update patterns.
4
+ metadata:
5
+ version: "0.0.1"
6
+ ---
7
+
8
+ # Documentation Guidance
9
+
10
+ ## Docs.md Purpose
11
+
12
+ `Docs.md` is the **authoritative technical reference** for implemented work. It serves as:
13
+
14
+ - The most comprehensive documentation of what was implemented
15
+ - A standalone technical reference for engineers
16
+ - The source of truth for understanding how the implementation works
17
+ - The basis for generating project-specific documentation
18
+ - Documentation that persists as the go-to reference
19
+
20
+ **Docs.md is NOT**: A list of documentation changes or a changelog.
21
+
22
+ ## Docs.md Template
23
+
24
+ ```markdown
25
+ # [Work Title]
26
+
27
+ ## Overview
28
+
29
+ [Comprehensive description of what was implemented, its purpose, and the problem it solves]
30
+
31
+ ## Architecture and Design
32
+
33
+ ### High-Level Architecture
34
+ [Architectural overview, system components, data flow]
35
+
36
+ ### Design Decisions
37
+ [Key design choices made during implementation and rationale]
38
+
39
+ ### Integration Points
40
+ [How this implementation integrates with existing systems]
41
+
42
+ ## User Guide
43
+
44
+ ### Prerequisites
45
+ [What users need before using this implementation]
46
+
47
+ ### Basic Usage
48
+ [Step-by-step guide for common use cases with examples]
49
+
50
+ ### Advanced Usage
51
+ [Complex scenarios, customization options, power-user features]
52
+
53
+ ## API Reference
54
+
55
+ ### Key Components
56
+ [Document reusable components that other code will call]
57
+
58
+ ### Configuration Options
59
+ [Available settings and their effects]
60
+
61
+ ## Testing
62
+
63
+ ### How to Test
64
+ [How to exercise the implementation as a human user]
65
+
66
+ ### Edge Cases
67
+ [Known edge cases and how they're handled]
68
+
69
+ ## Limitations and Future Work
70
+
71
+ [Known limitations, planned improvements, out-of-scope items]
72
+ ```
73
+
74
+ ## What to Include
75
+
76
+ Focus Docs.md on information not easily discovered from code:
77
+
78
+ | Include | Why |
79
+ |---------|-----|
80
+ | Design decisions and rationale | Not visible in code |
81
+ | Architecture and integration points | High-level view |
82
+ | User-facing behavior and usage patterns | User perspective |
83
+ | How to test/exercise as a human | Verification guidance |
84
+ | Migration paths and compatibility | Operational knowledge |
85
+ | Reusable components for other code | API surface |
86
+ | Edge cases and limitations | Gotchas users should know |
87
+
88
+ ## What NOT to Include
89
+
90
+ Avoid duplicating information already in code or other artifacts:
91
+
92
+ | Exclude | Why |
93
+ |---------|-----|
94
+ | Code reproduction | Already in the code |
95
+ | Every function/class | Internal details |
96
+ | Exhaustive API docs | Over-documentation |
97
+ | Test coverage checklists | In PRs and tests |
98
+ | Acceptance criteria verification | In implementation artifacts |
99
+ | Project doc updates list | In PR description |
100
+ | Unnecessary code examples | Only when essential |
101
+
102
+ ## Project Documentation Updates
103
+
104
+ When updating README, CHANGELOG, guides, or API docs, follow these principles:
105
+
106
+ ### Style Matching (CRITICAL)
107
+
108
+ **STUDY FIRST**: Before updating any project documentation:
109
+ 1. Read multiple existing entries/sections
110
+ 2. Understand the style, length, and detail level
111
+ 3. Match it precisely
112
+
113
+ ### CHANGELOG Discipline
114
+
115
+ - Create **ONE entry** for the work, not multiple sub-entries
116
+ - Follow existing format exactly (bullet style, sentence case, etc.)
117
+ - Group related changes into a single coherent entry
118
+
119
+ **Example:**
120
+ ```markdown
121
+ ## [Unreleased]
122
+
123
+ ### Added
124
+ - Skills-based architecture for implementation workflow (#164)
125
+ ```
126
+
127
+ ### README Discipline
128
+
129
+ - Match section length and detail level of surrounding sections
130
+ - Don't expand sections beyond the existing style
131
+ - Keep additions proportional to the significance of the change
132
+
133
+ ### When Uncertain
134
+
135
+ Err on the side of **LESS detail** in project docs. Docs.md contains the comprehensive detail; users can ask for more if needed.
136
+
137
+ ## Surgical Change Discipline
138
+
139
+ - ONLY modify documentation files required to describe the completed work
140
+ - DO NOT format or rewrite implementation code sections
141
+ - DO NOT introduce unrelated documentation updates or cleanups
142
+ - DO NOT remove historical context without explicit direction
143
+ - When unsure if a change is purely documentation, pause and ask
144
+
145
+ ## Documentation Depth by Workflow Mode
146
+
147
+ | Mode | Docs.md Depth | Project Docs |
148
+ |------|---------------|--------------|
149
+ | full | Comprehensive with all sections | Full updates |
150
+ | minimal | Essential information only | Minimal updates |
151
+ | custom | Per Custom Workflow Instructions | As specified |
152
+
153
+ ## Quality Checklist
154
+
155
+ Before completing documentation:
156
+
157
+ - [ ] Docs.md covers all sections relevant to the implementation
158
+ - [ ] Design decisions and rationale documented
159
+ - [ ] User-facing behavior explained with examples
160
+ - [ ] No code reproduction or over-documentation
161
+ - [ ] Project docs match existing style
162
+ - [ ] CHANGELOG has single coherent entry
163
+ - [ ] README changes proportional to significance
@@ -0,0 +1,196 @@
1
+ ---
2
+ name: paw-git-operations
3
+ description: Shared git mechanics for PAW activity skills including branch naming conventions, strategy-based branching logic, and selective staging discipline.
4
+ metadata:
5
+ version: "0.0.1"
6
+ ---
7
+
8
+ # Git Operations
9
+
10
+ ## Branch Naming Conventions
11
+
12
+ | Branch Type | Pattern | Example |
13
+ |-------------|---------|---------|
14
+ | Phase branch | `<target>_phase[N]` | `feature/auth-system_phase1` |
15
+ | Multi-phase | `<target>_phase[M-N]` | `feature/auth-system_phase1-3` |
16
+ | Planning branch | `<target>_plan` | `feature/auth-system_plan` |
17
+ | Docs branch | `<target>_docs` | `feature/auth-system_docs` |
18
+
19
+ **Target branch** is the feature branch from WorkflowContext.md (e.g., `feature/auth-system`).
20
+
21
+ ## Strategy-Based Branching Logic
22
+
23
+ ### PRs Strategy
24
+
25
+ Create intermediate branches for each workflow stage; push and create PRs for review.
26
+
27
+ **Phase work:**
28
+ 1. Check current branch: `git branch --show-current`
29
+ 2. If not on correct phase branch:
30
+ - Checkout target branch: `git checkout <target>`
31
+ - Set upstream if not set: `git branch --set-upstream-to=<remote>/<target>`
32
+ - Pull latest: `git pull`
33
+ - Create phase branch from target: `git checkout -b <target>_phase[N]`
34
+ 3. Verify: `git branch --show-current`
35
+ 4. Implement on phase branch, commit locally
36
+ 5. Push: `git push -u <remote> <target>_phase[N]`
37
+ 6. Create PR: `<target>_phase[N]` → `<target>`
38
+
39
+ **Planning work:**
40
+ 1. Checkout target branch: `git checkout <target>`
41
+ 2. Set upstream if not set: `git branch --set-upstream-to=<remote>/<target>`
42
+ 3. Pull latest: `git pull`
43
+ 4. Create planning branch: `git checkout -b <target>_plan`
44
+ 5. Commit planning artifacts
45
+ 6. Push: `git push -u <remote> <target>_plan`
46
+ 7. Create PR: `<target>_plan` → `<target>`
47
+
48
+ **Docs work:**
49
+ 1. Checkout target branch: `git checkout <target>`
50
+ 2. Set upstream if not set: `git branch --set-upstream-to=<remote>/<target>`
51
+ 3. Pull latest: `git pull`
52
+ 4. Create docs branch: `git checkout -b <target>_docs`
53
+ 5. Commit documentation
54
+ 6. Push: `git push -u <remote> <target>_docs`
55
+ 7. Create PR: `<target>_docs` → `<target>`
56
+
57
+ ### Local Strategy
58
+
59
+ Work directly on target branch; no intermediate branches or PRs between stages.
60
+
61
+ **All work:**
62
+ 1. Check current branch: `git branch --show-current`
63
+ 2. If not on target branch:
64
+ - Checkout target: `git checkout <target>`
65
+ - Set upstream if not set: `git branch --set-upstream-to=<remote>/<target>`
66
+ - Pull latest: `git pull`
67
+ 3. Verify: `git branch --show-current`
68
+ 4. Implement, commit to target branch
69
+ 5. Push: `git push <remote> <target>`
70
+ 6. Skip intermediate PR creation
71
+
72
+ ## Selective Staging Discipline
73
+
74
+ **CRITICAL**: Never use `git add .` or `git add -A`. Always stage files explicitly.
75
+
76
+ ### Standard Staging
77
+
78
+ ```bash
79
+ # Stage specific files
80
+ git add <file1> <file2> <file3>
81
+
82
+ # Verify staged changes before commit
83
+ git diff --cached
84
+ ```
85
+
86
+ ### PAW Artifact Staging
87
+
88
+ Before staging `.paw/` files, check if artifact tracking is disabled:
89
+
90
+ ```bash
91
+ # Check for .gitignore in work directory
92
+ if [ -f ".paw/work/<feature-slug>/.gitignore" ]; then
93
+ # Tracking disabled - skip .paw/ artifacts
94
+ git add <non-paw-files-only>
95
+ else
96
+ # Tracking enabled - stage all changed files
97
+ git add <all-changed-files>
98
+ fi
99
+ ```
100
+
101
+ **Why**: Users can disable artifact tracking via `.gitignore`. Respect this by checking before staging `.paw/` files.
102
+
103
+ ## Branch Verification
104
+
105
+ Before every commit, verify you're on the expected branch:
106
+
107
+ ```bash
108
+ # Get current branch
109
+ git branch --show-current
110
+
111
+ # Expected patterns by work type:
112
+ # - Phase work (prs): *_phase[N] or *_phase[M-N]
113
+ # - Planning (prs): *_plan
114
+ # - Docs (prs): *_docs
115
+ # - Any work (local): <target> (no suffix)
116
+ ```
117
+
118
+ **If on wrong branch**: STOP immediately. Do not commit. Switch to correct branch first.
119
+
120
+ ## Pre-Commit Checklist
121
+
122
+ 1. ✓ Verify current branch matches expected pattern
123
+ 2. ✓ Stage only related files (no `git add .`)
124
+ 3. ✓ Check `.paw/work/<slug>/.gitignore` before staging `.paw/` artifacts
125
+ 4. ✓ Review staged changes: `git diff --cached`
126
+ 5. ✓ Commit with descriptive message
127
+
128
+ ## Phase PR Creation
129
+
130
+ After implementation review passes, create Phase PR (PRs strategy only).
131
+
132
+ ### Push and Create PR
133
+
134
+ ```bash
135
+ # 1. Verify on phase branch
136
+ git branch --show-current # Should be <target>_phase[N]
137
+
138
+ # 2. Push branch
139
+ git push -u <remote> <target>_phase[N]
140
+
141
+ # 3. Create PR via gh CLI
142
+ gh pr create \
143
+ --base <target> \
144
+ --head <target>_phase[N] \
145
+ --title "[<Work Title>] Phase <N>: <description>" \
146
+ --body "<PR body>"
147
+ ```
148
+
149
+ ### PR Title Format
150
+
151
+ `[<Work Title>] Phase <N>: <one-sentence description>`
152
+
153
+ Example: `[Auth System] Phase 1: Add JWT token validation`
154
+
155
+ ### PR Body Scaling
156
+
157
+ **Simple phases**:
158
+ ```
159
+ Phase <N>: <one-sentence objective>
160
+
161
+ 🐾 Generated with [PAW](https://github.com/lossyrob/phased-agent-workflow)
162
+ ```
163
+
164
+ **Complex phases**:
165
+ ```
166
+ ## Summary
167
+ <Key changes and approach>
168
+
169
+ ## Design Decisions
170
+ <Noteworthy decisions made>
171
+
172
+ ## Reviewer Notes
173
+ <Items for reviewer attention>
174
+
175
+ 🐾 Generated with [PAW](https://github.com/lossyrob/phased-agent-workflow)
176
+ ```
177
+
178
+ ### Post-PR Actions
179
+
180
+ 1. Capture PR URL from `gh pr create` output
181
+ 2. Update ImplementationPlan.md with PR link in phase notes
182
+ 3. Report PR URL to user
183
+
184
+ ### PR Update Policy
185
+
186
+ After a PR is opened, post progress updates as **PR comments**, not modifications to the PR body. PR body modifications require explicit user request.
187
+
188
+ ### Reply Format (PR Comments)
189
+
190
+ When replying to review comments:
191
+
192
+ ```
193
+ **🐾 PAW 🤖:**
194
+
195
+ [What was changed and commit hash reference]
196
+ ```
@@ -0,0 +1,178 @@
1
+ ---
2
+ name: paw-impl-review
3
+ description: Implementation review activity skill for PAW workflow. Reviews implementation for quality, adds documentation, and returns structured verdict.
4
+ metadata:
5
+ version: "0.0.1"
6
+ ---
7
+
8
+ # Implementation Review
9
+
10
+ > **Execution Context**: This skill runs in a **subagent** session, delegated by the PAW orchestrator. Return structured feedback (pass/fail + issues) to the orchestrator—do not make orchestration decisions or perform git operations.
11
+
12
+ Review implementation changes for quality and maintainability. Acts as quality gate between implementation and PR creation.
13
+
14
+ > **Reference**: Follow Core Implementation Principles from `paw-workflow` skill.
15
+
16
+ ## Capabilities
17
+
18
+ - Review implementation for quality and maintainability
19
+ - Add documentation/docstrings to implementation
20
+ - Verify `paw-implement` addressed PR comments correctly
21
+ - Return structured verdict (pass/fail + issues)
22
+
23
+ ## Role: Maintainability
24
+
25
+ Focus on ensuring code is well-documented, readable, and maintainable.
26
+
27
+ **Responsibilities:**
28
+ - Review code for clarity, readability, necessity
29
+ - Question design decisions and identify unnecessary code
30
+ - Generate docstrings and code comments
31
+ - Make small refactors (remove unused parameters, simplify)
32
+ - Return structured verdict for orchestrator
33
+
34
+ **Not responsibilities** (handled elsewhere):
35
+ - Writing functional code or tests (`paw-implement`)
36
+ - Push branches or create PRs (PAW agent / `paw-git-operations`)
37
+ - Merging PRs (human responsibility)
38
+
39
+ ## Review Philosophy
40
+
41
+ Act as a critical PR reviewer, not just a documentation pass:
42
+
43
+ - Question whether code should exist as-is
44
+ - Identify unused parameters, dead code, over-engineering
45
+ - Check for code duplication across changed files
46
+ - Verify tests exist for new functionality
47
+
48
+ **Small refactors** (do yourself): Remove unused parameters, dead code, extract duplicate utilities
49
+ **Large refactors** (coordinate): Restructuring, major changes → return `blocked` with reason, specific changes needed, and evidence (file:line references, test output)
50
+
51
+ ## Project Instructions Adherence
52
+
53
+ Discover and enforce project-specific coding conventions from instruction files.
54
+
55
+ **Discovery**: Search for instruction files at repo root and `.github/`:
56
+ - `AGENTS.md`, `.github/AGENTS.md`
57
+ - `.github/copilot-instructions.md`
58
+ - `*.agent.md` in project root or `agents/` directory
59
+ - `.cursor/rules` or similar convention files
60
+
61
+ **Extract and verify**:
62
+ - Required commands (lint, build, test) → run them, must pass
63
+ - Coding patterns and conventions → verify in changed code
64
+ - Project-specific standards → check adherence
65
+
66
+ **Enforcement**: Non-adherence is a **blocking issue**, not a suggestion. If instructions specify a required command, failure = BLOCKED.
67
+
68
+ ## Desired End State
69
+
70
+ After review, the PAW agent receives:
71
+ - Pass/fail determination
72
+ - For passing reviews: confirmation ready for PR creation (with optional polish suggestions)
73
+ - For failing reviews: specific issues, what needs to change, and whether Implementer rework is required
74
+
75
+ ## Review Process
76
+
77
+ ### Plan Completeness Check (CRITICAL)
78
+
79
+ **Before reviewing code quality**, verify the implementation covers all plan items:
80
+
81
+ 1. **Read ImplementationPlan.md** for the current phase
82
+ 2. **Extract all planned changes**: files to modify, features to add, components to update
83
+ 3. **Compare against actual changes**: `git diff` against the plan checklist
84
+ 4. **Flag gaps**: If the plan says "update services A, B, C" but only A and B changed → BLOCKED
85
+
86
+ **This check catches partial implementations** where the agent implements some items but forgets others.
87
+
88
+ ### Initial Phase Review
89
+
90
+ **Required context**:
91
+ - Implementation changes via `git diff` or `git log`
92
+ - ImplementationPlan.md requirements for comparison
93
+
94
+ **Review focus**:
95
+ - Code clarity, readability, project conventions
96
+ - Code necessity: unused parameters, dead code, duplication
97
+ - Tests exist and pass (REQUIRED)
98
+
99
+ **Allowed improvements**:
100
+ - Add docstrings to new functions/classes
101
+ - Add inline comments for complex logic
102
+ - Small refactors (remove unused parameters, simplify)
103
+ - **Do NOT modify core functional logic**
104
+
105
+ **Constraints**:
106
+ - Commit improvements with clear messages
107
+ - Do NOT push or create PRs (orchestrator handles this)
108
+
109
+ ### Review Comment Verification
110
+
111
+ When reviewing Implementer's response to PR comments:
112
+
113
+ **Required context**:
114
+ - Implementer's commits present locally
115
+ - All PR comments and threads
116
+
117
+ **Verification**:
118
+ - Review Implementer's commits against the comments
119
+ - Run tests (REQUIRED):
120
+ - If tests fail due to reviewer changes → fix them
121
+ - If tests fail due to Implementer's code → return `blocked`
122
+ - If functional code changed but tests not updated → BLOCKER
123
+
124
+ **Constraints**:
125
+ - Add improvements if needed (documentation, polish)
126
+ - Do NOT push (orchestrator handles this)
127
+
128
+ ## Quality Checklist
129
+
130
+ - [ ] **Plan completeness verified**: All phase items from ImplementationPlan.md implemented
131
+ - [ ] Project instruction files discovered and reviewed
132
+ - [ ] Required commands from instructions pass (lint, build, test)
133
+ - [ ] Changes follow documented coding conventions
134
+ - [ ] All tests pass
135
+ - [ ] Reviewed for code necessity and duplication
136
+ - [ ] Docstrings added to public functions/classes
137
+ - [ ] No modifications to core functional logic
138
+ - [ ] Changes committed locally (not pushed)
139
+
140
+ ## Completion Response
141
+
142
+ Return structured feedback to PAW agent:
143
+
144
+ **PASS**: Implementation meets quality criteria, ready for PR
145
+ - Confirm tests pass
146
+ - Note any non-blocking observations
147
+ - Commits made (documentation, polish)
148
+
149
+ **BLOCKED**: Implementation needs rework
150
+ - List blocking issues with file:line references
151
+ - Specify what Implementer needs to change
152
+ - Note **incomplete plan items** (e.g., "Phase 2 says update services A, B, C but only A implemented")
153
+ - Note test failures or missing tests
154
+ - Note violations of project instruction conventions
155
+
156
+ > **After returning PASS**: The PAW orchestrator will handle push/PR creation (via `paw-git-operations`) and then invoke `paw-transition`. This skill does NOT push or create PRs—it only returns the verdict.
157
+
158
+ ### Response Format
159
+
160
+ ```
161
+ ## Review Result: [PASS|BLOCKED]
162
+
163
+ ### Summary
164
+ [One-sentence summary of review outcome]
165
+
166
+ ### Tests
167
+ - Status: [PASS|FAIL]
168
+ - [Details if relevant]
169
+
170
+ ### Commits Made
171
+ - [List of commits added during review, if any]
172
+
173
+ ### Issues Found
174
+ [For BLOCKED only: specific issues requiring Implementer attention]
175
+
176
+ ### Notes for Reviewer
177
+ [Optional: items flagged for human PR reviewer attention]
178
+ ```
@@ -0,0 +1,153 @@
1
+ ---
2
+ name: paw-implement
3
+ description: Implementation activity skill for PAW workflow. Executes plan phases with code changes, documentation phases, and PR review comment handling. One phase per invocation.
4
+ metadata:
5
+ version: "0.0.1"
6
+ ---
7
+
8
+ # Implementation
9
+
10
+ > **Execution Context**: This skill runs **directly** in the PAW session (not a subagent), preserving user interactivity for course-correction during implementation.
11
+
12
+ Execute implementation plan phases by making code changes, running verification, and committing locally. Operates one phase per invocation by default.
13
+
14
+ > **Reference**: Follow Core Implementation Principles from `paw-workflow` skill.
15
+
16
+ ## Critical: Preserve Existing Changes
17
+
18
+ **DO NOT modify, revert, or stage existing uncommitted changes** in the working directory. The user may have in-progress work they haven't committed yet.
19
+
20
+ - Before making changes, check `git status` to understand current state
21
+ - Only stage files YOU modified for the current phase
22
+ - Never use `git add .` or `git add -A` (can capture unrelated changes)
23
+ - If existing uncommitted changes conflict with plan work, report as `blocked` and ask user how to proceed
24
+
25
+ ## Capabilities
26
+
27
+ - Execute one or more plan phases based on delegation instructions
28
+ - Make focused code changes with appropriate verification (tests, lint) per repository norms
29
+ - Execute documentation phases (create/update Docs.md, update project documentation)
30
+ - Address PR review comments on implementation work (load `paw-review-response` for mechanics)
31
+ - Handle non-linear requests (e.g., "adjust implementation to match updated spec") when delegated by PAW agent
32
+
33
+ ## Role: Forward Momentum
34
+
35
+ Focus on making changes work and getting automated verification passing.
36
+
37
+ **Responsibilities:**
38
+ - Implement plan phases with functional code
39
+ - Run automated verification (tests, linting, type checking)
40
+ - Address PR review comments by making code changes
41
+ - Update ImplementationPlan.md with progress
42
+ - Commit functional changes locally
43
+
44
+ **Not responsibilities** (handled by `paw-impl-review`):
45
+ - Docstrings or code comments
46
+ - Code formatting or style polish
47
+ - Opening Phase PRs
48
+ - Replying to PR review comments
49
+
50
+ ## Implementation Philosophy
51
+
52
+ Plans are carefully designed, but reality can be messy:
53
+
54
+ - Follow the plan's intent while adapting to what you find
55
+ - Implement each phase fully before moving to the next
56
+ - Verify your work makes sense in the broader codebase context
57
+ - Update checkboxes in the plan as you complete sections
58
+
59
+ When things don't match the plan exactly, think about why and communicate clearly. The plan is your guide, but your judgment matters too.
60
+
61
+ ## Blocking Behavior
62
+
63
+ When plan conflicts with codebase reality or leaves critical gaps: **STOP immediately**. No TODO comments, placeholders, or speculative code. Return status `blocked` with specific blockers and what would resolve each.
64
+
65
+ **Exception**: If delegation explicitly allows degraded verification for non-critical checks (e.g., flaky tests, optional linters), report as `⚠️ warning` and proceed.
66
+
67
+ ## Execution Contexts
68
+
69
+ ### Initial Phase Development
70
+
71
+ **Desired end state**: Phase implemented, tests passing, changes committed locally on correct branch
72
+
73
+ **Required context**:
74
+ - Current phase requirements from ImplementationPlan.md
75
+ - Codebase conventions and verification commands from CodeResearch.md
76
+ - Branch state per Review Strategy (load `paw-git-operations`)
77
+
78
+ **Constraints**:
79
+ - All verification commands from CodeResearch.md must pass before committing
80
+ - Update ImplementationPlan.md: mark phase complete in Phase Status (`- [ ]` → `- [x]`)
81
+ - Commit locally with descriptive message
82
+ - **DO NOT push** — `paw-impl-review` handles that
83
+
84
+ ### Documentation Phase Execution
85
+
86
+ **Desired end state**: Docs.md created/updated, project docs updated (if warranted), docs build passing
87
+
88
+ **Required context**:
89
+ - Load `paw-docs-guidance` utility skill for templates and conventions
90
+ - Docs build command from CodeResearch.md (if framework discovered)
91
+
92
+ **Constraints**:
93
+ - Use same branch strategy and review flow as code phases
94
+ - Verify docs build passes before committing
95
+ - **DO NOT push** — `paw-impl-review` handles that
96
+
97
+ ### PR Review Comment Response
98
+
99
+ **Desired end state**: All review comments addressed with focused commits, ready for verification
100
+
101
+ **Required context**:
102
+ - PR metadata (URL or number) from delegation—if absent, return `blocked`
103
+ - PR type determined from PR description/metadata (Phase PR vs Final PR)
104
+ - Load `paw-review-response` for commit/reply mechanics
105
+ - Pull latest to include any reviewer commits
106
+
107
+ **Constraints**:
108
+ - One focused commit per comment group (related comments addressed together)
109
+ - Complete each group fully (changes + commit) before starting next
110
+ - **DO NOT push** — `paw-impl-review` verifies and pushes
111
+
112
+ ## Branching and Commits
113
+
114
+ > **Reference**: Load `paw-git-operations` skill for branch naming, commit mechanics, and selective staging.
115
+
116
+ ## Resuming Work
117
+
118
+ If the plan has existing checkmarks:
119
+ - Trust that completed work is done
120
+ - Pick up from the first unchecked item
121
+ - Verify previous work only if something seems off
122
+
123
+ ## Artifact Update Discipline
124
+
125
+ - Only update checkboxes for work actually completed in current session
126
+ - DO NOT mark phases complete preemptively
127
+ - Preserve prior notes; append new summaries
128
+ - Limit edits to sections affected by current phase
129
+ - Re-running same phase should produce no additional plan changes
130
+
131
+ ## Quality Checklist
132
+
133
+ ### Initial Phase Implementation
134
+
135
+ - [ ] All automated success criteria green
136
+ - [ ] Phase Status checkbox marked complete in ImplementationPlan.md
137
+ - [ ] Changes committed locally (NOT pushed)
138
+
139
+ ### PR Review Comment Response
140
+
141
+ - [ ] All automated checks passing
142
+ - [ ] Each comment group addressed with focused commit
143
+ - [ ] ImplementationPlan.md updated with "Addressed Review Comments:" section
144
+ - [ ] All commits local (NOT pushed)
145
+
146
+ ## Completion Response
147
+
148
+ Report back:
149
+ - Phase(s) completed and brief summary
150
+ - Verification results (tests, lint)
151
+ - Branch name and commit hash(es)
152
+ - Any items requiring user decision or review attention
153
+ - Status: `complete` or `blocked` (with specific blockers)