SUBAGENTS.md.tmpl

skills/cowork-project-architect/templates/SUBAGENTS.md.tmpl

1. ---
2. type: subagents
3. scope: project-root
4. tags:
5. - claude
6. - subagents
7. - reference
8. created: [YYYY-MM-DD]
9. updated: [YYYY-MM-DD]
10. ---
12. <!--
13. SUBAGENTS.md: Sub-agent definitions for this project.
15. Only generate this file if the project actually uses sub-agents.
16. Most small/medium projects don't need them.
18. Sub-agents are warranted when:
19. - Tasks split cleanly into specialist roles
20. - The project benefits from forced separation of concerns (e.g., a hygiene
21. step that strips bias from prompts before they hit a research agent)
22. - The project runs long agentic loops where a sub-agent prevents context bloat
23. -->
25. # [PROJECT NAME]: Sub-agents
27. This project uses [N] sub-agents. Dispatch them per the rules below.
29. ## When to use a sub-agent vs. doing it inline
31. [2-3 sentences on the principle for this project. E.g., "Use a sub-agent when the task requires a clean context (no prior conversation bias), or when the task will produce >2k tokens of intermediate work that doesn't need to live in the main thread."]
33. ---
35. ## [SUBAGENT-NAME]
37. **Role:** [One sentence. What this agent does and nothing else.]
39. **Model:** [Opus / Sonnet / Haiku], because [why]. See [[MODEL_GUIDE]] for project-wide model rules.
41. **When to dispatch:**
42. - [Trigger condition]
43. - [Trigger condition]
45. **Do NOT dispatch when:**
46. - [Anti-trigger]
48. **Prompt structure:**
50. ```
51. [Required opening: role statement, scope]
53. [Required context: what the agent needs to know]
55. Task: [The specific task being delegated]
57. Constraints:
58. - [Constraint]
59. - [Constraint]
61. Return format:
62. [Exact format the agent should respond in]
63. ```
65. **Returns:** [What the calling context should expect to receive. Format and shape.]
67. **Example dispatch:**
69. > [A real-shaped invocation showing what calling this agent looks like in practice.]
71. ---
73. ## [SUBAGENT-NAME]
75. [Same structure as above.]
77. ---
79. ## Sub-agent chain patterns
81. <!-- If sub-agents are commonly chained, document the pattern here. -->
83. ### Pattern: [NAME]
85. **Sequence:** [Agent A] → [Agent B] → [Agent C]
87. **When:** [Trigger for this whole chain]
89. **Hand-off rules:**
90. - [Agent A] returns [X]; [Agent B] consumes [X] as [Y]
91. - [Agent B] returns [Z]; [Agent C] consumes [Z]
93. **Failure modes to watch for:**
94. - [What goes wrong and how to detect it]