Unified Implement Skill - Implementation Plan
For Claude: REQUIRED SUB-SKILL: Use /wrangler:implement to execute this plan task-by-task.
Goal: Create unified /wrangler:implement skill that replaces both executing-plans and subagent-driven-development with autonomous, scope-agnostic implementation workflow.
Architecture: Slash command entry → Scope parser → Task executor with subagent dispatch → Automatic code review & fixes → Final verification → Completion options
Tech Stack:
- Markdown skill file (follows wrangler skill format)
- Slash command (
.md file in commands/)
- Integration with existing skills (TDD, verification, code review, finishing-branch)
- MCP tools for issue loading (issues_get, issues_list)
Implementation tracking: See MCP issues (project: implement-skill-unification)
Architecture Overview
Core Components
Scope Parser
- Input: User command string or conversation context
- Formats: spec files, plan files, issue IDs/ranges, context inference
- Output: Normalized task list
Task Executor
- Dispatches general-purpose subagent per task
- Enforces TDD via subagent prompt
- Collects TDD compliance certification
- Manages task dependencies
Code Review Automation
- Auto-dispatch code-reviewer subagent after each task
- Parse review feedback (Critical/Important/Minor)
- Auto-dispatch fix subagents for issues
- Retry logic with escalation (2 attempts max)
Blocker Detection
- Unclear requirements → immediate escalation
- Fix failures (2x) → escalation with "flummoxed" signal
- Missing dependencies → escalation
- Test failures after fixes → escalation
Verification & Completion
- Final test suite run
- Requirements checklist verification
- TDD compliance aggregation
- Integration with finishing-a-development-branch skill
Scope Format Handling
| Input Format |
Example |
Parsing Strategy |
| Spec file |
spec-auth.md |
Load from .wrangler/specifications/, extract linked issues or parse inline tasks |
| Plan file |
plan-refactor.md |
Load from plans/, extract task list |
| Single issue |
issue #42 or issue 42 |
Load via MCP issues_get |
| Issue range |
issues 5-7 or issues 5,6,7 |
Load via MCP issues_list with filtering |
| Context inference |
(none) |
Scan last 5 user messages for file/issue references |
Autonomous Execution Model
Key principle: No user checkpoints unless genuinely blocked
Stops only for:
- Unclear requirements (cannot infer intent)
- Flummoxed agents (2 fix failures on same issue)
- Missing dependencies (cannot proceed)
- Explicit user intervention requests
Does NOT stop for:
- Test failures (auto-fixes via subagent)
- Code review feedback (auto-fixes Critical/Important)
- Task completion (continues to next task automatically)
- Batch boundaries (no artificial checkpoints)
Integration Points
Required sub-skills:
test-driven-development - Subagents MUST follow TDDverification-before-completion - Final verificationrequesting-code-review - Code review templatefinishing-a-development-branch - Completion options
MCP tools:
issues_get - Load single issueissues_list - Load issue rangesissues_update - Optional: mark complete
Replaces:
executing-plans (hard deprecation - delete file)subagent-driven-development (hard deprecation - delete file)
Task Breakdown Summary
- Create implement skill foundation - Skill file, frontmatter, overview, scope parser
- Add task executor workflow - Subagent dispatch, TDD enforcement, dependency handling
- Add code review automation - Auto-review, feedback parsing, fix dispatch, retry logic
- Add blocker detection - Escalation conditions, flummoxed detection
- Add verification & completion - Final checks, TDD aggregation, finishing-branch integration
- Create slash command -
/wrangler:implement entry point
- Add examples & documentation - Usage examples, red flags, integration notes
- Delete deprecated skills - Remove executing-plans and subagent-driven-development
- Update references - Fix any skills referencing deprecated skills
- Verification & testing - Test with various scope formats, verify autonomous behavior
Design Decisions
Why Slash Command Entry?
- Discoverability: Users can find it via
/wrangler: prefix
- Explicit invocation: Clear when autonomous mode activates
- Clean syntax: No confusion about skill vs manual execution
Why Auto-Fix Instead of Stop-and-Ask?
- Efficiency: Most code review issues are straightforward fixes
- Maintains flow: Stopping after every task breaks momentum
- Safety net: 2-attempt limit prevents infinite loops
- Escalation path: User gets involved when truly needed
Why Merge Two Skills?
- Single mental model: One skill for all implementation
- Eliminates choice paralysis: No "which skill?" questions
- Simplifies maintenance: One workflow to update
- Reduces duplication: Both skills had 80% overlap
Why Hard Deprecation?
- Clean break: No confusion about which to use
- Prevents regressions: Can't accidentally use old workflow
- Forces migration: Ensures everyone benefits from improvements
- Simpler codebase: Less code to maintain
Success Criteria
1---2name: plans-bacchus-labs-wrangler3description: > For Claude: REQUIRED SUB-SKILL: Use /wrangler:implement to execute this plan task-by-task.4---5
6# Unified Implement Skill - Implementation Plan
7
8> **For Claude:** REQUIRED SUB-SKILL: Use /wrangler:implement to execute this plan task-by-task.
9
10**Goal:** Create unified `/wrangler:implement` skill that replaces both `executing-plans` and `subagent-driven-development` with autonomous, scope-agnostic implementation workflow.
11
12**Architecture:** Slash command entry → Scope parser → Task executor with subagent dispatch → Automatic code review & fixes → Final verification → Completion options
13
14**Tech Stack:**
15- Markdown skill file (follows wrangler skill format)
16- Slash command (`.md` file in `commands/`)
17- Integration with existing skills (TDD, verification, code review, finishing-branch)
18- MCP tools for issue loading (issues_get, issues_list)
19
20**Implementation tracking:** See MCP issues (project: implement-skill-unification)
21
22---
23
24## Architecture Overview
25
26### Core Components
27
281. **Scope Parser**
29 - Input: User command string or conversation context
30 - Formats: spec files, plan files, issue IDs/ranges, context inference
31 - Output: Normalized task list
32
332. **Task Executor**
34 - Dispatches general-purpose subagent per task
35 - Enforces TDD via subagent prompt
36 - Collects TDD compliance certification
37 - Manages task dependencies
38
393. **Code Review Automation**
40 - Auto-dispatch code-reviewer subagent after each task
41 - Parse review feedback (Critical/Important/Minor)
42 - Auto-dispatch fix subagents for issues
43 - Retry logic with escalation (2 attempts max)
44
454. **Blocker Detection**
46 - Unclear requirements → immediate escalation
47 - Fix failures (2x) → escalation with "flummoxed" signal
48 - Missing dependencies → escalation
49 - Test failures after fixes → escalation
50
515. **Verification & Completion**
52 - Final test suite run
53 - Requirements checklist verification
54 - TDD compliance aggregation
55 - Integration with finishing-a-development-branch skill
56
57### Scope Format Handling
58
59| Input Format | Example | Parsing Strategy |
60|--------------|---------|------------------|
61| Spec file | `spec-auth.md` | Load from `.wrangler/specifications/`, extract linked issues or parse inline tasks |
62| Plan file | `plan-refactor.md` | Load from `plans/`, extract task list |
63| Single issue | `issue #42` or `issue 42` | Load via MCP issues_get |
64| Issue range | `issues 5-7` or `issues 5,6,7` | Load via MCP issues_list with filtering |
65| Context inference | (none) | Scan last 5 user messages for file/issue references |
66
67### Autonomous Execution Model
68
69**Key principle:** No user checkpoints unless genuinely blocked
70
71**Stops only for:**
72- Unclear requirements (cannot infer intent)
73- Flummoxed agents (2 fix failures on same issue)
74- Missing dependencies (cannot proceed)
75- Explicit user intervention requests
76
77**Does NOT stop for:**
78- Test failures (auto-fixes via subagent)
79- Code review feedback (auto-fixes Critical/Important)
80- Task completion (continues to next task automatically)
81- Batch boundaries (no artificial checkpoints)
82
83### Integration Points
84
85**Required sub-skills:**
86- `test-driven-development` - Subagents MUST follow TDD
87- `verification-before-completion` - Final verification
88- `requesting-code-review` - Code review template
89- `finishing-a-development-branch` - Completion options
90
91**MCP tools:**
92- `issues_get` - Load single issue
93- `issues_list` - Load issue ranges
94- `issues_update` - Optional: mark complete
95
96**Replaces:**
97- `executing-plans` (hard deprecation - delete file)
98- `subagent-driven-development` (hard deprecation - delete file)
99
100---
101
102## Task Breakdown Summary
103
1041. **Create implement skill foundation** - Skill file, frontmatter, overview, scope parser
1052. **Add task executor workflow** - Subagent dispatch, TDD enforcement, dependency handling
1063. **Add code review automation** - Auto-review, feedback parsing, fix dispatch, retry logic
1074. **Add blocker detection** - Escalation conditions, flummoxed detection
1085. **Add verification & completion** - Final checks, TDD aggregation, finishing-branch integration
1096. **Create slash command** - `/wrangler:implement` entry point
1107. **Add examples & documentation** - Usage examples, red flags, integration notes
1118. **Delete deprecated skills** - Remove executing-plans and subagent-driven-development
1129. **Update references** - Fix any skills referencing deprecated skills
11310. **Verification & testing** - Test with various scope formats, verify autonomous behavior
114
115---
116
117## Design Decisions
118
119### Why Slash Command Entry?
120
121- **Discoverability:** Users can find it via `/wrangler:` prefix
122- **Explicit invocation:** Clear when autonomous mode activates
123- **Clean syntax:** No confusion about skill vs manual execution
124
125### Why Auto-Fix Instead of Stop-and-Ask?
126
127- **Efficiency:** Most code review issues are straightforward fixes
128- **Maintains flow:** Stopping after every task breaks momentum
129- **Safety net:** 2-attempt limit prevents infinite loops
130- **Escalation path:** User gets involved when truly needed
131
132### Why Merge Two Skills?
133
134- **Single mental model:** One skill for all implementation
135- **Eliminates choice paralysis:** No "which skill?" questions
136- **Simplifies maintenance:** One workflow to update
137- **Reduces duplication:** Both skills had 80% overlap
138
139### Why Hard Deprecation?
140
141- **Clean break:** No confusion about which to use
142- **Prevents regressions:** Can't accidentally use old workflow
143- **Forces migration:** Ensures everyone benefits from improvements
144- **Simpler codebase:** Less code to maintain
145
146---
147
148## Success Criteria
149
150- [ ] `/wrangler:implement` command exists and loads skill
151- [ ] Skill handles all scope formats (spec, plan, issue, range, context)
152- [ ] Subagents dispatched per task with TDD enforcement
153- [ ] Code review automatic after each task
154- [ ] Auto-fix for Critical/Important issues (max 2 attempts)
155- [ ] Escalates to user only for genuine blockers
156- [ ] Final verification runs before completion
157- [ ] Integrates with finishing-a-development-branch
158- [ ] executing-plans deleted
159- [ ] subagent-driven-development deleted
160- [ ] All references updated
161- [ ] Tested with various scenarios
majiayu000