1---
2name: subagent-driven-development
3description: Use when executing implementation plans with independent tasks in the current session
4---
5 
6# Subagent-Driven Development
7 
8Execute plan by dispatching fresh subagent per task, with two-stage review after each: spec compliance review first, then code quality review.
9 
10**Why subagents:** You delegate tasks to specialized agents with isolated context. By precisely crafting their instructions and context, you ensure they stay focused and succeed at their task. They should never inherit your session's context or history — you construct exactly what they need. This also preserves your own context for coordination work.
11 
12**Core principle:** Fresh subagent per task + two-stage review (spec then quality) = high quality, fast iteration
13 
14**Continuous execution:** Do not pause to check in with your human partner between tasks. Execute all tasks from the plan without stopping. The only reasons to stop are: BLOCKED status you cannot resolve, ambiguity that genuinely prevents progress, or all tasks complete. "Should I continue?" prompts and progress summaries waste their time — they asked you to execute the plan, so execute it.
15 
16## When to Use
17 
18```dot
19digraph when_to_use {
20    "Have implementation plan?" [shape=diamond];
21    "Tasks mostly independent?" [shape=diamond];
22    "Stay in this session?" [shape=diamond];
23    "subagent-driven-development" [shape=box];
24    "executing-plans" [shape=box];
25    "Manual execution or brainstorm first" [shape=box];
26 
27    "Have implementation plan?" -> "Tasks mostly independent?" [label="yes"];
28    "Have implementation plan?" -> "Manual execution or brainstorm first" [label="no"];
29    "Tasks mostly independent?" -> "Stay in this session?" [label="yes"];
30    "Tasks mostly independent?" -> "Manual execution or brainstorm first" [label="no - tightly coupled"];
31    "Stay in this session?" -> "subagent-driven-development" [label="yes"];
32    "Stay in this session?" -> "executing-plans" [label="no - parallel session"];
33}
34```
35 
36**vs. Executing Plans (parallel session):**
37- Same session (no context switch)
38- Fresh subagent per task (no context pollution)
39- Two-stage review after each task: spec compliance first, then code quality
40- Faster iteration (no human-in-loop between tasks)
41 
42## The Process
43 
44```dot
45digraph process {
46    rankdir=TB;
47 
48    subgraph cluster_per_task {
49        label="Per Task";
50        "Dispatch implementer subagent (./implementer-prompt.md)" [shape=box];
51        "Implementer subagent asks questions?" [shape=diamond];
52        "Answer questions, provide context" [shape=box];
53        "Implementer subagent implements, tests, commits, self-reviews" [shape=box];
54        "Dispatch spec reviewer subagent (./spec-reviewer-prompt.md)" [shape=box];
55        "Spec reviewer subagent confirms code matches spec?" [shape=diamond];
56        "Implementer subagent fixes spec gaps" [shape=box];
57        "Dispatch code quality reviewer subagent (./code-quality-reviewer-prompt.md)" [shape=box];
58        "Code quality reviewer subagent approves?" [shape=diamond];
59        "Implementer subagent fixes quality issues" [shape=box];
60        "Mark task complete in TodoWrite" [shape=box];
61    }
62 
63    "Read plan, extract all tasks with full text, note context, create TodoWrite" [shape=box];
64    "More tasks remain?" [shape=diamond];
65    "Dispatch final code reviewer subagent for entire implementation" [shape=box];
66    "Use superpowers:finishing-a-development-branch" [shape=box style=filled fillcolor=lightgreen];
67 
68    "Read plan, extract all tasks with full text, note context, create TodoWrite" -> "Dispatch implementer subagent (./implementer-prompt.md)";
69    "Dispatch implementer subagent (./implementer-prompt.md)" -> "Implementer subagent asks questions?";
70    "Implementer subagent asks questions?" -> "Answer questions, provide context" [label="yes"];
71    "Answer questions, provide context" -> "Dispatch implementer subagent (./implementer-prompt.md)";
72    "Implementer subagent asks questions?" -> "Implementer subagent implements, tests, commits, self-reviews" [label="no"];
73    "Implementer subagent implements, tests, commits, self-reviews" -> "Dispatch spec reviewer subagent (./spec-reviewer-prompt.md)";
74    "Dispatch spec reviewer subagent (./spec-reviewer-prompt.md)" -> "Spec reviewer subagent confirms code matches spec?";
75    "Spec reviewer subagent confirms code matches spec?" -> "Implementer subagent fixes spec gaps" [label="no"];
76    "Implementer subagent fixes spec gaps" -> "Dispatch spec reviewer subagent (./spec-reviewer-prompt.md)" [label="re-review"];
77    "Spec reviewer subagent confirms code matches spec?" -> "Dispatch code quality reviewer subagent (./code-quality-reviewer-prompt.md)" [label="yes"];
78    "Dispatch code quality reviewer subagent (./code-quality-reviewer-prompt.md)" -> "Code quality reviewer subagent approves?";
79    "Code quality reviewer subagent approves?" -> "Implementer subagent fixes quality issues" [label="no"];
80    "Implementer subagent fixes quality issues" -> "Dispatch code quality reviewer subagent (./code-quality-reviewer-prompt.md)" [label="re-review"];
81    "Code quality reviewer subagent approves?" -> "Mark task complete in TodoWrite" [label="yes"];
82    "Mark task complete in TodoWrite" -> "More tasks remain?";
83    "More tasks remain?" -> "Dispatch implementer subagent (./implementer-prompt.md)" [label="yes"];
84    "More tasks remain?" -> "Dispatch final code reviewer subagent for entire implementation" [label="no"];
85    "Dispatch final code reviewer subagent for entire implementation" -> "Use superpowers:finishing-a-development-branch";
86}
87```
88 
89## Model Selection
90 
91Use the least powerful model that can handle each role to conserve cost and increase speed.
92 
93**Mechanical implementation tasks** (isolated functions, clear specs, 1-2 files): use a fast, cheap model. Most implementation tasks are mechanical when the plan is well-specified.
94 
95**Integration and judgment tasks** (multi-file coordination, pattern matching, debugging): use a standard model.
96 
97**Architecture, design, and review tasks**: use the most capable available model.
98 
99**Task complexity signals:**
100- Touches 1-2 files with a complete spec → cheap model
101- Touches multiple files with integration concerns → standard model
102- Requires design judgment or broad codebase understanding → most capable model
103 
104## Handling Implementer Status
105 
106Implementer subagents report one of four statuses. Handle each appropriately:
107 
108**DONE:** Proceed to spec compliance review.
109 
110**DONE_WITH_CONCERNS:** The implementer completed the work but flagged doubts. Read the concerns before proceeding. If the concerns are about correctness or scope, address them before review. If they're observations (e.g., "this file is getting large"), note them and proceed to review.
111 
112**NEEDS_CONTEXT:** The implementer needs information that wasn't provided. Provide the missing context and re-dispatch.
113 
114**BLOCKED:** The implementer cannot complete the task. Assess the blocker:
1151. If it's a context problem, provide more context and re-dispatch with the same model
1162. If the task requires more reasoning, re-dispatch with a more capable model
1173. If the task is too large, break it into smaller pieces
1184. If the plan itself is wrong, escalate to the human
119 
120**Never** ignore an escalation or force the same model to retry without changes. If the implementer said it's stuck, something needs to change.
121 
122## Prompt Templates
123 
124- `./implementer-prompt.md` - Dispatch implementer subagent
125- `./spec-reviewer-prompt.md` - Dispatch spec compliance reviewer subagent
126- `./code-quality-reviewer-prompt.md` - Dispatch code quality reviewer subagent
127 
128## Example Workflow
129 
130```
131You: I'm using Subagent-Driven Development to execute this plan.
132 
133[Read plan file once: docs/superpowers/plans/feature-plan.md]
134[Extract all 5 tasks with full text and context]
135[Create TodoWrite with all tasks]
136 
137Task 1: Hook installation script
138 
139[Get Task 1 text and context (already extracted)]
140[Dispatch implementation subagent with full task text + context]
141 
142Implementer: "Before I begin - should the hook be installed at user or system level?"
143 
144You: "User level (~/.config/superpowers/hooks/)"
145 
146Implementer: "Got it. Implementing now..."
147[Later] Implementer:
148  - Implemented install-hook command
149  - Added tests, 5/5 passing
150  - Self-review: Found I missed --force flag, added it
151  - Committed
152 
153[Dispatch spec compliance reviewer]
154Spec reviewer: ✅ Spec compliant - all requirements met, nothing extra
155 
156[Get git SHAs, dispatch code quality reviewer]
157Code reviewer: Strengths: Good test coverage, clean. Issues: None. Approved.
158 
159[Mark Task 1 complete]
160 
161Task 2: Recovery modes
162 
163[Get Task 2 text and context (already extracted)]
164[Dispatch implementation subagent with full task text + context]
165 
166Implementer: [No questions, proceeds]
167Implementer:
168  - Added verify/repair modes
169  - 8/8 tests passing
170  - Self-review: All good
171  - Committed
172 
173[Dispatch spec compliance reviewer]
174Spec reviewer: ❌ Issues:
175  - Missing: Progress reporting (spec says "report every 100 items")
176  - Extra: Added --json flag (not requested)
177 
178[Implementer fixes issues]
179Implementer: Removed --json flag, added progress reporting
180 
181[Spec reviewer reviews again]
182Spec reviewer: ✅ Spec compliant now
183 
184[Dispatch code quality reviewer]
185Code reviewer: Strengths: Solid. Issues (Important): Magic number (100)
186 
187[Implementer fixes]
188Implementer: Extracted PROGRESS_INTERVAL constant
189 
190[Code reviewer reviews again]
191Code reviewer: ✅ Approved
192 
193[Mark Task 2 complete]
194 
195...
196 
197[After all tasks]
198[Dispatch final code-reviewer]
199Final reviewer: All requirements met, ready to merge
200 
201Done!
202```
203 
204## Advantages
205 
206**vs. Manual execution:**
207- Subagents follow TDD naturally
208- Fresh context per task (no confusion)
209- Parallel-safe (subagents don't interfere)
210- Subagent can ask questions (before AND during work)
211 
212**vs. Executing Plans:**
213- Same session (no handoff)
214- Continuous progress (no waiting)
215- Review checkpoints automatic
216 
217**Efficiency gains:**
218- No file reading overhead (controller provides full text)
219- Controller curates exactly what context is needed
220- Subagent gets complete information upfront
221- Questions surfaced before work begins (not after)
222 
223**Quality gates:**
224- Self-review catches issues before handoff
225- Two-stage review: spec compliance, then code quality
226- Review loops ensure fixes actually work
227- Spec compliance prevents over/under-building
228- Code quality ensures implementation is well-built
229 
230**Cost:**
231- More subagent invocations (implementer + 2 reviewers per task)
232- Controller does more prep work (extracting all tasks upfront)
233- Review loops add iterations
234- But catches issues early (cheaper than debugging later)
235 
236## Red Flags
237 
238**Never:**
239- Start implementation on main/master branch without explicit user consent
240- Skip reviews (spec compliance OR code quality)
241- Proceed with unfixed issues
242- Dispatch multiple implementation subagents in parallel (conflicts)
243- Make subagent read plan file (provide full text instead)
244- Skip scene-setting context (subagent needs to understand where task fits)
245- Ignore subagent questions (answer before letting them proceed)
246- Accept "close enough" on spec compliance (spec reviewer found issues = not done)
247- Skip review loops (reviewer found issues = implementer fixes = review again)
248- Let implementer self-review replace actual review (both are needed)
249- **Start code quality review before spec compliance is ✅** (wrong order)
250- Move to next task while either review has open issues
251 
252**If subagent asks questions:**
253- Answer clearly and completely
254- Provide additional context if needed
255- Don't rush them into implementation
256 
257**If reviewer finds issues:**
258- Implementer (same subagent) fixes them
259- Reviewer reviews again
260- Repeat until approved
261- Don't skip the re-review
262 
263**If subagent fails task:**
264- Dispatch fix subagent with specific instructions
265- Don't try to fix manually (context pollution)
266 
267## Integration
268 
269**Required workflow skills:**
270- **superpowers:using-git-worktrees** - Ensures isolated workspace (creates one or verifies existing)
271- **superpowers:writing-plans** - Creates the plan this skill executes
272- **superpowers:requesting-code-review** - Code review template for reviewer subagents
273- **superpowers:finishing-a-development-branch** - Complete development after all tasks
274 
275**Subagents should use:**
276- **superpowers:test-driven-development** - Subagents follow TDD for each task
277 
278**Alternative workflow:**
279- **superpowers:executing-plans** - Use for parallel session instead of same-session execution
280