Orchestrate
Coordinate substantial work across focused subagents while remaining available
to the user and retaining responsibility for the integrated result.
When to Use
- Use when a task has multiple independent research, review, or implementation lanes.
- Use when parallel work will materially reduce elapsed time or improve coverage.
- Use when a coordinator must synthesize several bounded outputs into one verified result.
Keep trivial tasks with the coordinator.
Workflow
- Decompose the task into distinct, bounded assignments with explicit outputs.
- Run narrow, read-only scouts in parallel with low reasoning effort and no
inherited conversation when the runtime supports those controls. Give each
scout all scoped context and evidence required to complete its assignment.
- Use medium reasoning effort for routine implementation and high reasoning
effort for difficult work.
- Give each subagent distinct ownership. Prevent overlapping assignments, and
instruct leaf workers not to delegate.
- Integrate the outputs, resolve conflicts, and verify the combined result.
- Keep approvals and externally consequential decisions with the user.
Examples
- For a repository-wide feature, assign non-overlapping agents to architecture
inspection, implementation, and test review, then integrate their findings
and run the final verification from the coordinator.
- For a research brief, assign independent sources or questions to read-only
scouts, reconcile disagreements, and keep the final judgment with the
coordinator.
Limitations
- Requires a runtime that exposes subagent or delegation tools; otherwise keep
the work with the coordinator.
- Delegation does not authorize file mutations, public actions, purchases, or
other consequential operations beyond the user's original scope.
- Parallel agents can add cost and coordination overhead, so use them only when
the task is substantial enough to benefit.
- The coordinator remains responsible for checking claims, changes, tests, and
the final answer.
Source: sickn33/agentic-awesome-skills → skills/orchestrate/SKILL.md
Also appears in: sickn33/agentic-awesome-skills/plugins/agentic-awesome-skills/skills/orchestrate/SKILL.md, sickn33/agentic-awesome-skills/plugins/agentic-awesome-skills-claude/skills/orchestrate/SKILL.md
1---2name: orchestrate-23description: Coordinate focused subagents on substantial work, keep their ownership non-overlapping, and integrate verified results. Use for large-scope Codex tasks; keep trivial work with the coordinator.4---5
6
7# Orchestrate
8
9Coordinate substantial work across focused subagents while remaining available
10to the user and retaining responsibility for the integrated result.
11
12## When to Use
13
14- Use when a task has multiple independent research, review, or implementation lanes.
15- Use when parallel work will materially reduce elapsed time or improve coverage.
16- Use when a coordinator must synthesize several bounded outputs into one verified result.
17
18Keep trivial tasks with the coordinator.
19
20## Workflow
21
221. Decompose the task into distinct, bounded assignments with explicit outputs.
232. Run narrow, read-only scouts in parallel with low reasoning effort and no
24 inherited conversation when the runtime supports those controls. Give each
25 scout all scoped context and evidence required to complete its assignment.
263. Use medium reasoning effort for routine implementation and high reasoning
27 effort for difficult work.
284. Give each subagent distinct ownership. Prevent overlapping assignments, and
29 instruct leaf workers not to delegate.
305. Integrate the outputs, resolve conflicts, and verify the combined result.
316. Keep approvals and externally consequential decisions with the user.
32
33## Examples
34
35- For a repository-wide feature, assign non-overlapping agents to architecture
36 inspection, implementation, and test review, then integrate their findings
37 and run the final verification from the coordinator.
38- For a research brief, assign independent sources or questions to read-only
39 scouts, reconcile disagreements, and keep the final judgment with the
40 coordinator.
41
42## Limitations
43
44- Requires a runtime that exposes subagent or delegation tools; otherwise keep
45 the work with the coordinator.
46- Delegation does not authorize file mutations, public actions, purchases, or
47 other consequential operations beyond the user's original scope.
48- Parallel agents can add cost and coordination overhead, so use them only when
49 the task is substantial enough to benefit.
50- The coordinator remains responsible for checking claims, changes, tests, and
51 the final answer.
52
53---
54
55**Source:** [`sickn33/agentic-awesome-skills`](https://github.com/sickn33/agentic-awesome-skills) → `skills/orchestrate/SKILL.md`
56
57**Also appears in:** `sickn33/agentic-awesome-skills/plugins/agentic-awesome-skills/skills/orchestrate/SKILL.md`, `sickn33/agentic-awesome-skills/plugins/agentic-awesome-skills-claude/skills/orchestrate/SKILL.md`
thedixitjain