返回 CodeWhale
SKILL.md
根目录 / crates / tui / assets / skills / delegate / SKILL.md
1 ---
2 name: delegate
3 description: Strategic delegation for multi-step coding, research, or verification work. Use when a task can be split into parent reasoning plus focused sub-agent execution through the agent tool.
4 metadata:
5 short-description: Delegate focused work to sub-agents
6 ---
7
8 # Delegate
9
10 Use sub-agents when they can do focused work in parallel while the parent keeps architectural judgment, integration, and final verification.
11
12 ## Keep vs Delegate
13
14 Keep in the parent:
15
16 - Understanding the user's actual request and constraints
17 - Architecture, security, product, and release-risk decisions
18 - Cross-module integration
19 - Final review, test interpretation, and user-facing summary
20
21 Delegate to sub-agents:
22
23 - Read-only exploration over a bounded file set
24 - Mechanical edits with a clear file ownership boundary
25 - Focused test or lint runs
26 - Boilerplate generation from an explicit spec
27 - Independent checks that can run while parent work continues
28
29 Do not delegate tiny one-step tasks, ambiguous product decisions, destructive operations without a clear acceptance criterion, or final verification.
30
31 ## Launch Focused Work
32
33 Use `agent` for a focused child run. Launch independent children together so they can run in parallel.
34
35 Prefer provider-neutral `model_strength` over hardcoded model ids. Children inherit the active model by default (`model_strength: "same"`), including `type: "scout"`, so pass `model_strength: "faster"` explicitly to get the cheaper same-family sibling for read-only exploration:
36
37 ```json
38 {
39 "name": "config_audit",
40 "prompt": "Inspect crates/tui/src/config.rs and crates/tui/src/settings.rs for duplicate model-default logic. Return file/line findings only; do not edit files.",
41 "type": "scout",
42 "model_strength": "faster",
43 "cwd": "."
44 }
45 ```
46
47 For code changes, give the child a precise write boundary and tell it not to revert unrelated edits. Keep implementation children capable with `model_strength: "same"`:
48
49 ```json
50 {
51 "name": "docs_patch",
52 "prompt": "Update only docs/configuration.md to document the new [statusline] keys. Match the surrounding style. Do not edit other files.",
53 "type": "builder",
54 "model_strength": "same",
55 "cwd": "."
56 }
57 ```
58
59 Use `fork_context: true` only when the child genuinely needs the current conversation prefix. Leave it omitted for fresh, narrower context.
60
61 ## Evaluate and Verify
62
63 Sub-agent outputs are self-reports. Re-check material claims before relying on them:
64
65 - Read changed files directly.
66 - Run the relevant tests locally.
67 - Inspect unexpected diffs before committing.
68 - Verify externally visible or destructive claims against source data.
69
70 ## Prompt Shape
71
72 A good delegation prompt includes:
73
74 - The exact task
75 - Files or modules owned by the child
76 - Files or behavior the child must not touch
77 - Expected output format
78 - Acceptance criteria
79
80 Weak prompt:
81
82 ```text
83 Fix the settings bug.
84 ```
85
86 Strong prompt:
87
88 ```text
89 Own only crates/tui/src/settings.rs and its tests. Preserve existing config key names. Add a regression test showing that provider-specific API key changes do not restart DeepSeek onboarding. Return the changed paths and test command output.
90 ```
91
91 lines MARKDOWN