intelligent-delegation
par hogpile
Search & ResearchRésumé
A 5-phase framework for reliable AI-to-AI task delegation, inspired by Google DeepMind's "Intelligent AI.
Description
Ce que fait ce skill
A 5-phase framework for AI-to-AI task delegation based on Google DeepMind's Intelligent AI Delegation paper (arXiv 2602.11865). Covers task tracking, sub-agent performance logging, automated output verification, fallback chains, and systematic agent selection via multi-axis scoring.
Replaces ad-hoc delegation with a structured, self-correcting system that logs outcomes, enforces contracts, and routes failures automatically rather than silently dropping them.
Quand l'utiliser
- Tracking background sub-agent tasks so none go unmonitored
- Choosing which sub-agent to assign a task based on past performance
- Automatically verifying sub-agent output before accepting results
- Recovering from delegation failures without manual intervention
- Scoring task complexity and reversibility to decide autonomy level
Exemple de workflow
User asks: delegate a data extraction task to a sub-agent and ensure results are valid
- 1
Score the task on 7 axes using score_task.py to determine autonomy level and agent choice
- 2
Write a task contract specifying expected output format, success criteria, and fallback steps
- 3
Log the task to TASKS.md and schedule a cron job to check completion
- 4
On completion, run verify_task.py to confirm file exists and passes structural checks
- 5
If verification fails, apply the fallback chain: retry with adjusted prompt, then try a different agent, then escalate
Verified output file at expected path, task marked COMPLETED in TASKS.md, outcome logged to agent-performance.md
Installation
clawhub install hogpile/intelligent-delegationExécute la commande CLI, ou colle l'instruction dans Claude Code, Cursor, Codex ou l'agent de code que tu utilises.