Skip to content
Skills Library
AI & agents · Automation

parallel-launch

/parallel-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.

Licence
Apache-2.0
Last repository change
28.09.2026

Assessment and security

90/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content25 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
3
Scripts
0
HTML comments
0
URLs in SKILL.md
0
Deprecation

No signals detected. Repository changed on 28.09.2026, without references to retired models or APIs.

Licence

Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.

What it does and when to use it

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.

Original author description

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.

How it works

  1. Contract
  2. Inputs
  3. Procedure
  4. Failure and recovery
  5. Output
Original SKILL.md headings · 36 lines, 700 words
SKILL.md preview
---
name: parallel-launch
description: '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.'
---

# Parallel launch

## Contract

| Field | Bound contract |
|---|---|
| Trigger | The request has multiple independent sub-tasks, cross-domain research, or parallelizable work. |
| Authority | Reversible local: writes only local targets explicitly named in the request or agent assignments; rollback is the captured prior state or exact undo operation. No remote mutation. |
| Side effect | Execute concurrent subagents and report composed output in chat. This orchestration creates no durable file of its own; delegated local changes remain confined to their named targets. |
| Done | Return composed output after completeness, consistency, accuracy, and scope review passes, or return each usable partial result with the exact gap and a targeted follow-up when an agent or review fails. |

## Inputs

Required: the request, its acceptance criteria, and enough source context for each sub-task to be executed without shared agent memory. For implementation, the request must name every local target that agents may change. Optional: output format, domain-specific agent choices, and ordering constraints.

## Procedure

1. Split the request by independent concern. A concern is independent only when it shares no mutable target and has no ordering dependency with another concern. Do not split work merely to increase agent count; if fewer than two concerns are independent, execute the request directly. Done when: the request is split into independent concerns or executed directly.
2. Put dependent concerns into later batches whose prerequisites are explicit. Before any mutation, list each batch's named local targets and capture the prior content or an exact undo operation for each target. Done when: dependent concerns are batched with prerequisites and prior states captured.
3. Give each concern to one task-specific agent with the complete context it needs, a bounded objective, named writable targets, acceptance criteria, and a concrete result format. Agents must not rely on memory or output from another concurrent agent. Done when: each concern has a self-contained agent brief.
4. Launch every concern in the current independent batch together in one concurrent dispatch. Wait for the batch to finish before dispatching work that depends on it. Done when: the batch is dispatched concurrently and dependent work is held.
5. Compose completed results. Merge non-conflicting outputs; reconcile overlaps against the request and report any unresolved trade-off rather than choosing without evidence. Done when: non-conflicting outputs are merged and overlaps are reconciled or reported.
6. Dispatch an independent reviewer with the original request, acceptance criteria, and composed result. Require it to check that every concern is covered, outputs do not contradict one another, claims are supported, and nothing exceeds scope; for implementation, also check specification compliance and code quality. Done when: the reviewer checks coverage, consistency, claims, and scope.
7. Return the composition only when review passes. If an agent or review fails, preserve usable results and return the exact gap plus one targeted follow-up that would close it; do not widen scope or invent missing evidence. Done when: the composition is returned on review pass, or the gap and follow-up are returned on failure.

## Failure and recovery
- Invalid partition: Shared mutable targets or hidden ordering make a batch unsafe. Do not launch that batch; repartition it into sequential work, or return `blocked` with the conflicting targets or dependency.
- Agent failure or incomplete result: Keep independently usable results unchanged and return `blocked` with the failed concern, observed error, missing acceptance criterion, and targeted follow-up.
- Conflicting results: Reconcile only from supplied evidence. If evidence cannot resolve the conflict, return `non-converged` with both positions, the unresolved fact, and the targeted check needed.
- Review failure: Do not present the composition as complete. Return `non-converged` with failed review criteria, usable partial results, and a targeted follow-up.
- Unauthorized or partial mutation: Stop further dispatch. Restore every changed local target from its captured prior content or exact undo operation, then report restoration status and any target that could not be restored. Never conceal an error or claim the done predicate holds.

## Output
One reviewed composition mapping each concern to its result with review passage recorded, or, on failure, usable partial results, terminal classification (`blocked` or `non-converged`), exact gap, observed error or conflict, restoration status, and one targeted follow-up.
Files in this package (3)
  • SKILL.md 5 KB
  • agents/openai.yaml 1 KB
  • references/delegation-scenarios.md 1 KB

Versions

RepositoryScoreLicence
This page · OutlineDriven/outline-driven-development90Apache-2.0
gosuda/dalcom-skills90Apache-2.0
OutlineDriven/odin-codex-plugin90Apache-2.0

Database snapshot: 13.11.2026 · package version ddab08ee5454 (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

  • 90partition-scopes-to-subagents
    Use when asked to partition non-overlapping scopes across subagents and drive each to completion. Not for concurrent independent sub-tasks: use parallel-launch.
  • 95partition-scopes-to-subagents
    Use when asked to partition non-overlapping scopes across subagents and drive each to completion. Not for concurrent independent sub-tasks: use parallel-launch.
  • 100agent-native
    Dispatch independent tasks to parallel workers or subagents without write collisions. Use when: running or planning agents in parallel, even two; check scopes before any launch.
  • 90task-init-fleet
    Orchestrator-workers variant of task-init that decomposes a complex brief into N parallel independent sub-tasks, each worker running a full task-init. Use when the brief contains N >= 3 logically independent work streams (e.g., REST->GraphQL migration across api-server +…
  • 100spawn-research-subagent
    Use when a research task splits into independent sub-questions or needs heavy reading across many sources. Fans the work out to parallel Haiku subagents. Not for single-fact lookups.
  • 100run-parallel
    [Orbit 2.4.4 Implementation] Independent tasks with parallel agents after an independence check. Use when: «параллельно», «одновременно несколько», «разгони задачи», «несколько агентов сразу»; parallel agents, run in parallel.

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.