multi-agent
/multi-agent
Coordinate authorized independent agent subtasks with explicit scopes, ownership and evidence. Use when delegation, a fleet, parallel agent work or result synthesis is requested or appropriate under host instructions; keep sequential or small work in one agent.
- Author
- wrongstack ↗
- Repository
- wrongstack ↗
- Licence
- MIT
- Last repository change
- 27.10.2026
Assessment and security
- Files in this package
- 3
- Scripts
- 0
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 27.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
What it does and when to use it
Coordinate authorized independent agent subtasks with explicit scopes, ownership and evidence. Use when delegation, a fleet, parallel agent work or result synthesis is requested or appropriate under host instructions; keep sequential or small work in one agent.
Original author description
Coordinate authorized independent agent subtasks with explicit scopes, ownership and evidence. Use when delegation, a fleet, parallel agent work or result synthesis is requested or appropriate under host instructions; keep sequential or small work in one agent.
How it works
- Selection card
- Overview
- Rules
- Workflow
- Brief template
- Review workflow
- Before returning
- Skills in scope
SKILL.md preview
--- name: multi-agent description: "Coordinate authorized independent agent subtasks with explicit scopes, ownership and evidence. Use when delegation, a fleet, parallel agent work or result synthesis is requested or appropriate under host instructions; keep sequential or small work in one agent." version: 2.1.1 required-capabilities: [filesystem.read] required-tools: [] optional-capabilities: [verification.run, web.research, filesystem.write, fleet.delegate] trigger: "delegation, a fleet, parallel agent work or result synthesis is requested or appropriate under host instructions; keep sequential or small work in one agent." metadata: routing-group: workflow --- # Multi-Agent Coordination ## Selection card - Task: Plan coordinated work when delegation is authorized. - Start: Identify the requested artifact, repository owner and acceptance criteria. - Finish: apply the acceptance checks below; report observed results and unresolved constraints. ## Overview Parallel agents improve independent attention and elapsed time, at the cost of handoff and synthesis. Use the host's actual limits and context-sharing rules. Separate contexts may still share files, build outputs, ports and databases. ## Rules 1. Follow the session's delegation authorization and available tools. A large task or this skill's availability alone does not authorize spawning agents. 2. Delegate independent work with exact boundaries and a definition of done. Keep decisions requiring the complete context with the coordinator. 3. Assign file/resource ownership before writes. Prefer parallel reading and separate edit scopes; serialize shared build, coverage, release and database writers unless their isolation is demonstrated. 4. Treat task brief and supplied artifacts as the worker's reliable context; inspect what the host actually inherits instead of assuming total amnesia or unlimited shared memory. 5. Validate every worker's completion status and evidence. Partial, cancelled and budget-exhausted outputs are coverage gaps, not successful checks. 6. Synthesize one outcome with deduplicated findings and material cross-scope effects; do not concatenate reports or fabricate unanimity. ## Workflow 1. Decide whether the work is independent enough to repay briefing and merge costs. Use one agent when a subtask depends on exploratory results. 2. Brief each worker: task, exact paths, raw inputs, constraints, write ownership, checks, return format, budget and stopping conditions. 3. Dispatch within host concurrency/resource limits. Keep useful coordinator work running while workers execute. 4. Handle dependency handoffs explicitly with versioned artifacts or paths. A worker should reread a shared file before editing when another writer changed it. 5. Check actual statuses, commands and changed files. Read relevant code for consequential findings and inspect integration conflicts. 6. Retry a failed worker only when a changed scope/input addresses the cause. Bound retries; respect cancellation and report remaining gaps. 7. Run the integration checks against the combined final state with one owner. ## Brief template ~~~ Task: <one concrete objective> Scope and ownership: <paths; read-only or permitted writes> Inputs and constraints: <artifacts; relevant host rules> Done when: <observable result> Verify with: <commands and expected behavior> Return: <outcome, files/lines, commands/results, limitations> Stop/escalate when: <dependency or scope decision> ~~~ ## Review workflow For the host's collab_debug path, load [the complete instructions](references/collab-debug.md) before dispatch. That workflow's actual tool contract governs roles and result handling. ## Before returning - [ ] Delegation authorization and host concurrency limits were verified and respected. - [ ] File and shared-resource write ownership was partitioned or serialized before dispatch. - [ ] Every worker completion status (including partial, cancelled, or timed-out runs) and evidence was reconciled. - [ ] Combined final state passed integration checks under a single owner, with any coverage gaps or conflicting observations reported. ## Skills in scope - `code-review` — independent review of worker-produced diffs and cross-module contract changes - `evidence-audit` — scoped proof-driven verification rounds across isolated subsystems - `refactor-planner` — dependency graph ordering and file-ownership partitioning before parallel execution - `audit-log` — inspection of worker session trajectories, tool calls, and failure traces - `output-standards` — deduplicated synthesis of multi-worker results into a unified report
Files in this package (3)
- SKILL.md 5 KB
- SKILL.save.md 2 KB
- references/collab-debug.md 4 KB
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · wrongstack/wrongstack | 90 | MIT |
| shibing624/agentica | 100 | Apache-2.0 |
| masihmoafi/skills | 100 | MIT |
| fengshao1227/ccg-workflow | 95 | MIT |
| GuYu-001/ccg-ros2-workflow | 95 | MIT |
| samchon/typia | 90 | MIT |
| systemfsoftware/systemfsoftware | 90 | Apache-2.0 |
Database snapshot: 13.11.2026 · package version c5288d1160fc (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 85agent-orchestrator
Plan and coordinate authorized multi-agent work with explicit dependencies, resource ownership, acceptance checks and integration. Use when the user requests delegation, or an authorized engineering workflow identifies independent tasks worth parallel execution and host policy… - 100orchestrate
This skill should be used when the user asks to 'orchestrate a task', 'break down work into parallel agents', 'coordinate subtasks', 'run agents in parallel', or mentions 'multi-agent'. Decomposes complex tasks into tracked subtasks, dispatches parallel subagents, and coordinates until completion. - 90multi-agent-collaboration
Coordinate multiple agents on implementation or research with isolated ownership, reproducible handoffs, independent checks, and evidence-based conflict resolution. Use when parallel or peer-agent work is requested or already in progress. - 100native-agent-swarms
Coordinate small teams of specialized Codex agents with bounded parallelism, explicit ownership, native messaging, and evidence-based synthesis. Use when two or more independent investigations, reviews, or implementation streams can run concurrently without sharing writable… - 90parallel-launch
Use when work splits into independent sub-tasks or cross-domain research. Not for durable scope ownership with per-scope review: use partition-scopes-to-subagents. - 90delegation-design
Break a bounded engineering request into independent, reviewable work units for a human team or an already available agent system. Use when parallel work is requested and the environment supports it.