subagent-delegationlisted

MANDATORY before starting any multi-part task, spawning any subagent, doing 3+ searches or tool calls, reading or summarizing long material, bulk-processing items, or independently checking important claims. Inspect the active subagent types first, route bounded work to the cheapest advertised suitable worker, prevent duplicate work, and keep synthesis in the main conversation.
alectivism/organization-ai-skills · ★ 0 · AI & Automation · score 75

Install: claude install-skill alectivism/organization-ai-skills

# Subagent Delegation Treat the main session as the orchestrator. Keep planning, user judgment, synthesis, and final drafting in the main conversation. Delegate bounded gathering, reading, checking, and repetitive work when doing so saves time, cost, or main-context space. Do not assume that a named agent, model, or reasoning level exists. ChatGPT Work uses hosted agents. Local Codex can also load custom TOML agents. The available catalog can differ by product, account, client, and release. ## Start with the active runtime Before spawning anything, inspect the collaboration or subagent tool schema shown in the session. Treat a role, model, and effort combination as available only when that schema advertises it. Apply this order: 1. Use `org-researcher`, `org-summarizer`, `org-bulk-worker`, `org-fact-checker`, or `org-verifier` when the runtime advertises those installed agents. Use `terra-ingest` for long-context ingestion and `sol-reviewer` for an adversarial read-only review when those installed agents are available. 2. Choose the specialist by task, not by model preference: - `org-researcher` is the default for multi-source and source-constrained research. - `org-summarizer` is the default for faithful compression of long notes, documents, threads, or transcripts. - `org-bulk-worker` is the default Luna route for high-volume, bounded, low-judgment work. - `org-fact-checker` is for a claim-by-claim check of a draft against sour