subagent-driven-development

Name: subagent-driven-development
Author: NeoLabHQ

911installs132forks

How to install subagent-driven-development?

neolabhq/context-engineering-kit/subagent-driven-development

$npx -y skills add NeoLabHQ/context-engineering-kit --skill subagent-driven-development

Installs into the current project.

Summary

Use when executing implementation plans with independent tasks in the current session or facing 3+ independent issues that can be investigated without shared state or dependencies - dispatches fresh subagent for each task with code review between tasks, enabling fast iteration wi

Files · 1

View on GitHub

SKILL.md

1	# Subagent-Driven Development
2
3	Create and execute plan by dispatching fresh subagent per task or issue, with code and output review after each or batch of tasks.
4
5	Core principle: Fresh subagent per task + review between or after tasks = high quality, fast iteration.
6
7	Executing Plans through agents:
8
9	- Same session (no context switch)
10	- Fresh subagent per task (no context pollution)
11	- Code review after each or batch of task (catch issues early)
12	- Faster iteration (no human-in-loop between tasks)
13
14	## Supported types of execution
15
16	### Sequential Execution
17
18	When you have a tasks or issues that are related to each other, and they need to be executed in order, investigating or modifying them sequentially is the best way to go.
19
20	Dispatch one agent per task or issue. Let it work sequentially. Review the output and code after each task or issue.
21
22	When to use:
23
24	- Tasks are tightly coupled
25	- Tasks should be executed in order
26
27	### Parallel Execution
28
29	When you have multiple unrelated tasks or issues (different files, different subsystems, different bugs), investigatin or modifying them sequentially wastes time. Each task or investigation is independent and can happen in parallel.
30
31	Dispatch one agent per independent problem domain. Let them work concurrently.
32
33	When to use:
34
35	- Tasks are mostly independent
36	- Overral review can be done after all tasks are completed
37
38	## Sequential Execution Process
39
40	### 1. Load Plan
41
42	Read plan file, create TodoWrite with all tasks.
43
44	### 2. Execute Task with Subagent
45
46	For each task:
47
48	Dispatch fresh subagent:
49
50	```
51	Task tool (general-purpose):
52	description: "Implement Task N: [task name]"
53	prompt: \|
54	You are implementing Task N from [plan-file].
55
56	Read that task carefully. Your job is to:
57	1. Implement exactly what the task specifies
58	2. Write tests (following TDD if task says to)
59	3. Verify implementation works
60	4. Commit your work
61	5. Report back
62
63	Work from: [directory]
64
65	Report: What you implemented, what you tested, test results, files changed, any issues
66	```
67
68	Subagent reports back with summary of work.
69
70	### 3. Review Subagent's Work
71
72	Dispatch code-reviewer subagent:
73
74	```
75	Task tool (superpowers:code-reviewer):
76	Use template at requesting-code-review/code-reviewer.md
77
78	WHAT_WAS_IMPLEMENTED: [from subagent's report]
79	PLAN_OR_REQUIREMENTS: Task N from [plan-file]
80	BASE_SHA: [commit before task]
81	HEAD_SHA: [current commit]
82	DESCRIPTION: [task summary]
83	```
84
85	Code reviewer returns: Strengths, Issues (Critical/Important/Minor), Assessment
86
87	### 4. Apply Review Feedback
88
89	If issues found:
90
91	- Fix Critical issues immediately
92	- Fix Important issues before next task
93	- Note Minor issues
94
95	Dispatch follow-up subagent if needed:
96
97	```
98	"Fix issues from code review: [list issues]"
99	```
100
101	### 5. Mark Complete, Next Task
102
103	- Mark task as completed in TodoWrite
104	- Move to next task
105	- Repeat steps 2-5
106
107	### 6. Final Review
108
109	After all tasks complete, dispatch final code-reviewer:
110
111	- Reviews entire implementation
112	- Checks all plan requirements met
113	- Validates overall architecture
114
115	### 7. Complete Development
116
117	After final review passes:
118
119	- Announce: "I'm using the finishing-a-development-branch skill to complete this work."
120	- REQUIRED SUB-SKILL: Use superpowers:finishing-a-development-branch
121	- Follow that skill to verify tests, present options, execute choice
122
123	### Example Workflow
124
125	```
126	You: I'm using Subagent-Driven Development to execute this plan.
127
128	[Load plan, create TodoWrite]
129
130	Task 1: Hook installation script
131
132	[Dispatch implementation subagent]
133	Subagent: Implemented install-hook with tests, 5/5 passing
134
135	[Get git SHAs, dispatch code-reviewer]
136	Reviewer: Strengths: Good test coverage. Issues: None. Ready.
137
138	[Mark Task 1 complete]
139
140	Task 2: Recovery modes
141
142	[Dispatch implementation subagent]
143	Subagent: Added verify/repair, 8/8 tests passing
144
145	[Dispatch code-reviewer]
146	Reviewer: Strengths: Solid. Issues (Important): Missing progress reporting
147
148	[Dispatch fix subagent]
149	Fix subagent: Added progress every 100 conversations
150
151	[Verify fix, mark Task 2 complete]
152
153	...
154
155	[After all tasks]
156	[Dispatch final code-reviewer]
157	Final reviewer: All requirements met, ready to merge
158
159	Done!
160	```
161
162	### Red Flags
163
164	Never:
165
166	- Skip code review between tasks
167	- Proceed with unfixed Critical issues
168	- Dispatch multiple implementation subagents in parallel (conflicts)
169	- Implement without reading plan task
170
171	If subagent fails task:
172
173	- Dispatch fix subagent with specific instructions
174	- Don't try to fix manually (context pollution)
175
176	## Parallel Execution Process
177
178	Load plan, review critically, execute tasks in batches, report for review betwe

People who liked this also viewed

6 picks

Type

subagent-driven-development

911installs132forks

View on GitHub

How to install subagent-driven-development?

neolabhq/context-engineering-kit/subagent-driven-development

$npx -y skills add NeoLabHQ/context-engineering-kit --skill subagent-driven-development

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Subagent-Driven Development
2
3	Create and execute plan by dispatching fresh subagent per task or issue, with code and output review after each or batch of tasks.
4
5	Core principle: Fresh subagent per task + review between or after tasks = high quality, fast iteration.
6
7	Executing Plans through agents:
8
9	- Same session (no context switch)
10	- Fresh subagent per task (no context pollution)
11	- Code review after each or batch of task (catch issues early)
12	- Faster iteration (no human-in-loop between tasks)
13
14	## Supported types of execution
15
16	### Sequential Execution
17
18	When you have a tasks or issues that are related to each other, and they need to be executed in order, investigating or modifying them sequentially is the best way to go.
19
20	Dispatch one agent per task or issue. Let it work sequentially. Review the output and code after each task or issue.
21
22	When to use:
23
24	- Tasks are tightly coupled
25	- Tasks should be executed in order
26
27	### Parallel Execution
28
29	When you have multiple unrelated tasks or issues (different files, different subsystems, different bugs), investigatin or modifying them sequentially wastes time. Each task or investigation is independent and can happen in parallel.
30
31	Dispatch one agent per independent problem domain. Let them work concurrently.
32
33	When to use:
34
35	- Tasks are mostly independent
36	- Overral review can be done after all tasks are completed
37
38	## Sequential Execution Process
39
40	### 1. Load Plan
41
42	Read plan file, create TodoWrite with all tasks.
43
44	### 2. Execute Task with Subagent
45
46	For each task:
47
48	Dispatch fresh subagent:
49
50	```
51	Task tool (general-purpose):
52	description: "Implement Task N: [task name]"
53	prompt: \|
54	You are implementing Task N from [plan-file].
55
56	Read that task carefully. Your job is to:
57	1. Implement exactly what the task specifies
58	2. Write tests (following TDD if task says to)
59	3. Verify implementation works
60	4. Commit your work
61	5. Report back
62
63	Work from: [directory]
64
65	Report: What you implemented, what you tested, test results, files changed, any issues
66	```
67
68	Subagent reports back with summary of work.
69
70	### 3. Review Subagent's Work
71
72	Dispatch code-reviewer subagent:
73
74	```
75	Task tool (superpowers:code-reviewer):
76	Use template at requesting-code-review/code-reviewer.md
77
78	WHAT_WAS_IMPLEMENTED: [from subagent's report]
79	PLAN_OR_REQUIREMENTS: Task N from [plan-file]
80	BASE_SHA: [commit before task]
81	HEAD_SHA: [current commit]
82	DESCRIPTION: [task summary]
83	```
84
85	Code reviewer returns: Strengths, Issues (Critical/Important/Minor), Assessment
86
87	### 4. Apply Review Feedback
88
89	If issues found:
90
91	- Fix Critical issues immediately
92	- Fix Important issues before next task
93	- Note Minor issues
94
95	Dispatch follow-up subagent if needed:
96
97	```
98	"Fix issues from code review: [list issues]"
99	```
100
101	### 5. Mark Complete, Next Task
102
103	- Mark task as completed in TodoWrite
104	- Move to next task
105	- Repeat steps 2-5
106
107	### 6. Final Review
108
109	After all tasks complete, dispatch final code-reviewer:
110
111	- Reviews entire implementation
112	- Checks all plan requirements met
113	- Validates overall architecture
114
115	### 7. Complete Development
116
117	After final review passes:
118
119	- Announce: "I'm using the finishing-a-development-branch skill to complete this work."
120	- REQUIRED SUB-SKILL: Use superpowers:finishing-a-development-branch
121	- Follow that skill to verify tests, present options, execute choice
122
123	### Example Workflow
124
125	```
126	You: I'm using Subagent-Driven Development to execute this plan.
127
128	[Load plan, create TodoWrite]
129
130	Task 1: Hook installation script
131
132	[Dispatch implementation subagent]
133	Subagent: Implemented install-hook with tests, 5/5 passing
134
135	[Get git SHAs, dispatch code-reviewer]
136	Reviewer: Strengths: Good test coverage. Issues: None. Ready.
137
138	[Mark Task 1 complete]
139
140	Task 2: Recovery modes
141
142	[Dispatch implementation subagent]
143	Subagent: Added verify/repair, 8/8 tests passing
144
145	[Dispatch code-reviewer]
146	Reviewer: Strengths: Solid. Issues (Important): Missing progress reporting
147
148	[Dispatch fix subagent]
149	Fix subagent: Added progress every 100 conversations
150
151	[Verify fix, mark Task 2 complete]
152
153	...
154
155	[After all tasks]
156	[Dispatch final code-reviewer]
157	Final reviewer: All requirements met, ready to merge
158
159	Done!
160	```
161
162	### Red Flags
163
164	Never:
165
166	- Skip code review between tasks
167	- Proceed with unfixed Critical issues
168	- Dispatch multiple implementation subagents in parallel (conflicts)
169	- Implement without reading plan task
170
171	If subagent fails task:
172
173	- Dispatch fix subagent with specific instructions
174	- Don't try to fix manually (context pollution)
175
176	## Parallel Execution Process
177
178	Load plan, review critically, execute tasks in batches, report for review betwe

People who liked this also viewed

6 picks

Type