orchestrate-tasks

Estrelas11

Forks0

Atualizado29 de maio de 2026 às 18:59

Fan large tasks across parallel agents with planner/worker/verifier roles and structured handoffs. Trigger: 'parallelize', 'fan out', 'orchestrate', or when plan-slice produces 3+ independent slices.

Instalação

Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.

Fonte

aryaminus

aryaminus/controlkeel

Abrir repositório GitHub Ver repositórios do creator

Download

Ocupações relacionadasSOC

Baseado na classificação ocupacional SOC

Desenvolvedores de softwareInformática e Matemática·SOC 15-1252

Explorador de arquivos

2 arquivos

SKILL.md

readonly

name	orchestrate-tasks
description	Fan large tasks across parallel agents with planner/worker/verifier roles and structured handoffs. Trigger: 'parallelize', 'fan out', 'orchestrate', or when plan-slice produces 3+ independent slices.
when_to_use	Activate ONLY when a task is too large for a single agent and has independent parallelizable slices. Do NOT use for single-agent tasks.
argument-hint	[goal or slice plan to orchestrate]
disable-model-invocation	true
license	Apache-2.0
compatibility	["opencode-native","claude-standalone","cursor-native","codex"]
metadata	{"author":"controlkeel","version":"1.1","category":"orchestration","ck_mcp_tools":["ck_route","ck_delegate","ck_budget","ck_review_submit","ck_finding","ck_git_diff","ck_memory_record","ck_rollback"],"related_skills":["plan-slice"]}

Orchestrate Tasks

Fan large tasks across parallel agents with planner/worker/verifier roles. Adapted from Cursor's orchestrate plugin for CK governance.

Do NOT use when

Single-agent task (just use controlkeel-governance)
Task has no parallelizable slices
Budget is tight (orchestration is expensive — multiple delegations)

Principles

Planners own scope, not code — decompose, publish tasks, read handoffs, decide next. No file edits.
Workers are isolated — one task, one agent, one scope. Talk up through handoffs only.
Verifiers check independently — read worker output, verify against acceptance criteria.
State on disk — ck_memory_record and git, no long-running agent state.
Continuous convergence — each cycle produces progress. No progress → escalate.

Workflow

Check ck_budget — orchestration costs multiple delegations. Stop if insufficient.
Decompose into tasks (or use existing plan-slice output). Each task: name, type (worker/verifier/subplanner), scoped goal, acceptance criteria, dependencies, paths.
Record plan via ck_memory_record (type: decision, tags: orchestration-plan).
Submit plan via ck_review_submit (review_type: plan). Wait for approval before spawning.
Execute loop:
- Select ready tasks (dependencies satisfied, not started)
- Check budget before each delegation
- Delegate: ck_route → ck_delegate
- Collect handoffs
- Verify completed tasks
- ck_rollback checkpoint after each success
- Decide: new tasks, retry failures, or completion
Synthesize: ck_git_diff for total change, parallel-review if large.

Failure recovery

Worker failure → record ck_finding, retry once, escalate if retry fails
Budget exhaustion → checkpoint state, stop, record for resumption
Verifier failure → worker output not trusted, fix or escalate

Output

Approved orchestration plan
Delegated tasks with handoffs
ck_finding for failures
ck_rollback checkpoints
Synthesized result

Reference

Handoff format

Mais deste repositório

mesmo repositório

bounded-loop

aryaminus/controlkeel

Run objective, repeatable improvement work under an immutable verifier and hard iteration, cost, time, and no-progress limits. Use for approved experiments with an automated metric and sandboxed worker.

2026-07-1711

communication-style

aryaminus/controlkeel

Write concise, calm technical responses whose structure matches the decision, evidence, or sequence being communicated. Use when drafting substantial plans, reviews, findings, documentation, or completion summaries.

2026-07-1711

controlkeel-governance

aryaminus/controlkeel

Operate inside a ControlKeel-governed session. Use this before code edits, shell execution, delegation, deploy work, or any task that needs CK validation, findings, budget, proof, or routing context.

2026-07-1711

cost-optimization

aryaminus/controlkeel

Keep a governed session within budget. Use this before long-running agent work, bulk processing, or any task where spend pressure could change the plan.

2026-07-1711

end-of-shift

aryaminus/controlkeel

Close a governed work session with validation, proof, findings, budget, digest, learning, and handoff checks. Use when the user asks to wrap up, stop for the day, or leave work ready for another agent.

2026-07-1711

false-confidence-test-audit

aryaminus/controlkeel

Audit tests that may pass without proving the claimed behavior. Use for periodic test-quality reviews or when coverage looks healthy but regressions still escape.

2026-07-1711

name	orchestrate-tasks
description	Fan large tasks across parallel agents with planner/worker/verifier roles and structured handoffs. Trigger: 'parallelize', 'fan out', 'orchestrate', or when plan-slice produces 3+ independent slices.
when_to_use	Activate ONLY when a task is too large for a single agent and has independent parallelizable slices. Do NOT use for single-agent tasks.
argument-hint	[goal or slice plan to orchestrate]
disable-model-invocation	true
license	Apache-2.0
compatibility	["opencode-native","claude-standalone","cursor-native","codex"]
metadata	{"author":"controlkeel","version":"1.1","category":"orchestration","ck_mcp_tools":["ck_route","ck_delegate","ck_budget","ck_review_submit","ck_finding","ck_git_diff","ck_memory_record","ck_rollback"],"related_skills":["plan-slice"]}

Orchestrate Tasks

Fan large tasks across parallel agents with planner/worker/verifier roles. Adapted from Cursor's orchestrate plugin for CK governance.

Do NOT use when

Single-agent task (just use controlkeel-governance)
Task has no parallelizable slices
Budget is tight (orchestration is expensive — multiple delegations)

Principles

Planners own scope, not code — decompose, publish tasks, read handoffs, decide next. No file edits.
Workers are isolated — one task, one agent, one scope. Talk up through handoffs only.
Verifiers check independently — read worker output, verify against acceptance criteria.
State on disk — ck_memory_record and git, no long-running agent state.
Continuous convergence — each cycle produces progress. No progress → escalate.

Workflow

Check ck_budget — orchestration costs multiple delegations. Stop if insufficient.
Decompose into tasks (or use existing plan-slice output). Each task: name, type (worker/verifier/subplanner), scoped goal, acceptance criteria, dependencies, paths.
Record plan via ck_memory_record (type: decision, tags: orchestration-plan).
Submit plan via ck_review_submit (review_type: plan). Wait for approval before spawning.
Execute loop:
- Select ready tasks (dependencies satisfied, not started)
- Check budget before each delegation
- Delegate: ck_route → ck_delegate
- Collect handoffs
- Verify completed tasks
- ck_rollback checkpoint after each success
- Decide: new tasks, retry failures, or completion
Synthesize: ck_git_diff for total change, parallel-review if large.

Failure recovery

Worker failure → record ck_finding, retry once, escalate if retry fails
Budget exhaustion → checkpoint state, stop, record for resumption
Verifier failure → worker output not trusted, fix or escalate

Output

Approved orchestration plan
Delegated tasks with handoffs
ck_finding for failures
ck_rollback checkpoints
Synthesized result

Reference

Handoff format