subagent-driven-development

Name: subagent-driven-development
Author: AlexAI-MCP

132stars24forks

How to install subagent-driven-development?

alexai-mcp/hermes-ccc/subagent-driven-development

$npx -y skills add AlexAI-MCP/hermes-CCC --skill subagent-driven-development

Installs into the current project.

Summary

Decompose Claude Code work into parallel subagent-friendly streams when the environment permits delegation. Use when a task has independent sidecar work, disjoint file ownership, or review/verification streams that can run in parallel without blocking the critical path.

Files · 1

View on GitHub

SKILL.md

1	# Subagent-Driven Development
2
3	## Purpose
4
5	- Shorten total task time by parallelizing independent work.
6	- Keep the critical path moving locally while sidecar work runs elsewhere.
7	- Prevent duplicated effort between contributors.
8	- Improve reliability by separating implementation, research, and verification concerns.
9	- Turn broad requests into explicit owned slices.
10
11	## Activation Signals
12
13	- Use this skill when the task has multiple independent questions.
14	- Use this skill when write scopes can be separated cleanly.
15	- Use this skill when verification can happen in parallel with implementation.
16	- Use this skill when one thread can map code while another edits disjoint files.
17	- Use this skill only if the environment allows subagent delegation.
18
19	## Non-Use Signals
20
21	- Do not use this skill when the user did not permit delegation.
22	- Do not use this skill when the next local action is blocked on the answer.
23	- Do not use this skill when the task is small enough to finish faster alone.
24	- Do not use this skill when the subtasks touch the same files and will conflict.
25	- Do not use this skill when the problem is still poorly framed.
26
27	## Core Principle
28
29	- Keep blocking work local.
30	- Delegate bounded sidecar work.
31	- Assign explicit ownership.
32	- Integrate results quickly.
33	- Do not re-do delegated work yourself.
34
35	## Decomposition Procedure
36
37	1. Identify the deliverable.
38	2. Identify the next blocking action.
39	3. Keep that blocking action local.
40	4. List other work that can advance in parallel.
41	5. Split the work by independent outputs or disjoint write sets.
42	6. Write one concrete prompt per subagent.
43	7. State ownership and constraints clearly.
44	8. Continue local progress immediately.
45
46	## Good Subtasks
47
48	- codebase mapping for a specific subsystem
49	- docs-backed verification
50	- isolated backend patch in owned files
51	- isolated frontend patch in owned files
52	- targeted regression test creation in a separate file family
53	- review of a completed patch
54
55	## Bad Subtasks
56
57	- "figure out the whole problem"
58	- "do whatever seems useful"
59	- work that overlaps the same file region
60	- urgent work needed for the next local decision
61	- tasks that duplicate what the main thread is already doing
62
63	## Prompt Design Rules
64
65	- State the task in one sentence first.
66	- Name the owned files or module boundary.
67	- Say that the subagent is not alone in the codebase.
68	- Say not to revert other changes.
69	- Define the expected final output.
70	- Include any known constraints or acceptance tests.
71
72	## Example Prompt
73
74	```text
75	Implement the parser fix in parser/tokenizer.py only. You are not alone in the codebase, so do not revert unrelated edits and adjust to concurrent changes if needed. Return the files changed and a short note on verification.
76	```
77
78	## Ownership Rules
79
80	- One write owner per file family whenever possible.
81	- Shared read-only context is fine.
82	- Shared write scope is a last resort.
83	- Verification agents should prefer read-only work unless asked to patch tests.
84
85	## Waiting Rules
86
87	- Do not wait immediately after delegation.
88	- Use the time to do non-overlapping local work.
89	- Wait only when integration is blocked on the result.
90	- Prefer longer waits over frequent polling.
91	- Close finished agents when they are no longer needed.
92
93	## Integration Procedure
94
95	1. Review the returned output.
96	2. Check whether the subtask stayed within scope.
97	3. Read the changed files or summary.
98	4. Integrate into the local branch carefully.
99	5. Reconcile any assumptions that conflict with local work.
100	6. Run the relevant validation.
101
102	## Conflict Management
103
104	- If two streams unexpectedly touch the same file, stop and replan.
105	- Do not blindly merge contradictory changes.
106	- Preserve the clearer ownership model after the replan.
107	- Prefer one integrator making final decisions.
108
109	## Verification Patterns
110
111	- implementation locally, review remotely
112	- implementation remotely, integration locally
113	- code mapping remotely, coding locally
114	- docs verification remotely, code changes locally
115
116	## Fallback When Subagents Are Not Allowed
117
118	- Use the same decomposition framework locally.
119	- Convert each subagent slice into a checklist item.
120	- Work the slices sequentially in priority order.
121	- Preserve explicit ownership even if ownership is just "current thread".
122
123	## Decision Rules
124
125	- Delegate only if the subtask is concrete and bounded.
126	- Keep ambiguity resolution local.
127	- Parallelize information gathering aggressively when it is

People who liked this also viewed

6 picks

Type

subagent-driven-development

132stars24forks

View on GitHub

How to install subagent-driven-development?

alexai-mcp/hermes-ccc/subagent-driven-development

$npx -y skills add AlexAI-MCP/hermes-CCC --skill subagent-driven-development

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Subagent-Driven Development
2
3	## Purpose
4
5	- Shorten total task time by parallelizing independent work.
6	- Keep the critical path moving locally while sidecar work runs elsewhere.
7	- Prevent duplicated effort between contributors.
8	- Improve reliability by separating implementation, research, and verification concerns.
9	- Turn broad requests into explicit owned slices.
10
11	## Activation Signals
12
13	- Use this skill when the task has multiple independent questions.
14	- Use this skill when write scopes can be separated cleanly.
15	- Use this skill when verification can happen in parallel with implementation.
16	- Use this skill when one thread can map code while another edits disjoint files.
17	- Use this skill only if the environment allows subagent delegation.
18
19	## Non-Use Signals
20
21	- Do not use this skill when the user did not permit delegation.
22	- Do not use this skill when the next local action is blocked on the answer.
23	- Do not use this skill when the task is small enough to finish faster alone.
24	- Do not use this skill when the subtasks touch the same files and will conflict.
25	- Do not use this skill when the problem is still poorly framed.
26
27	## Core Principle
28
29	- Keep blocking work local.
30	- Delegate bounded sidecar work.
31	- Assign explicit ownership.
32	- Integrate results quickly.
33	- Do not re-do delegated work yourself.
34
35	## Decomposition Procedure
36
37	1. Identify the deliverable.
38	2. Identify the next blocking action.
39	3. Keep that blocking action local.
40	4. List other work that can advance in parallel.
41	5. Split the work by independent outputs or disjoint write sets.
42	6. Write one concrete prompt per subagent.
43	7. State ownership and constraints clearly.
44	8. Continue local progress immediately.
45
46	## Good Subtasks
47
48	- codebase mapping for a specific subsystem
49	- docs-backed verification
50	- isolated backend patch in owned files
51	- isolated frontend patch in owned files
52	- targeted regression test creation in a separate file family
53	- review of a completed patch
54
55	## Bad Subtasks
56
57	- "figure out the whole problem"
58	- "do whatever seems useful"
59	- work that overlaps the same file region
60	- urgent work needed for the next local decision
61	- tasks that duplicate what the main thread is already doing
62
63	## Prompt Design Rules
64
65	- State the task in one sentence first.
66	- Name the owned files or module boundary.
67	- Say that the subagent is not alone in the codebase.
68	- Say not to revert other changes.
69	- Define the expected final output.
70	- Include any known constraints or acceptance tests.
71
72	## Example Prompt
73
74	```text
75	Implement the parser fix in parser/tokenizer.py only. You are not alone in the codebase, so do not revert unrelated edits and adjust to concurrent changes if needed. Return the files changed and a short note on verification.
76	```
77
78	## Ownership Rules
79
80	- One write owner per file family whenever possible.
81	- Shared read-only context is fine.
82	- Shared write scope is a last resort.
83	- Verification agents should prefer read-only work unless asked to patch tests.
84
85	## Waiting Rules
86
87	- Do not wait immediately after delegation.
88	- Use the time to do non-overlapping local work.
89	- Wait only when integration is blocked on the result.
90	- Prefer longer waits over frequent polling.
91	- Close finished agents when they are no longer needed.
92
93	## Integration Procedure
94
95	1. Review the returned output.
96	2. Check whether the subtask stayed within scope.
97	3. Read the changed files or summary.
98	4. Integrate into the local branch carefully.
99	5. Reconcile any assumptions that conflict with local work.
100	6. Run the relevant validation.
101
102	## Conflict Management
103
104	- If two streams unexpectedly touch the same file, stop and replan.
105	- Do not blindly merge contradictory changes.
106	- Preserve the clearer ownership model after the replan.
107	- Prefer one integrator making final decisions.
108
109	## Verification Patterns
110
111	- implementation locally, review remotely
112	- implementation remotely, integration locally
113	- code mapping remotely, coding locally
114	- docs verification remotely, code changes locally
115
116	## Fallback When Subagents Are Not Allowed
117
118	- Use the same decomposition framework locally.
119	- Convert each subagent slice into a checklist item.
120	- Work the slices sequentially in priority order.
121	- Preserve explicit ownership even if ownership is just "current thread".
122
123	## Decision Rules
124
125	- Delegate only if the subtask is concrete and bounded.
126	- Keep ambiguity resolution local.
127	- Parallelize information gathering aggressively when it is

People who liked this also viewed

6 picks

Type