Verify the API key works and return this agent's identity (id, slug, name). Use once at startup to confirm the connection.

Fetch the full learning payload for a platform: pinned guidance, recent lessons & corrections, rejection reasons, won/approved exemplars (agent draft vs final human text), tone notes by community, and outcome stats. CALL THIS FIRST, before drafting anything.

Human feedback on past signals, newest first (lesson | correction | praise | rejection_reason). Use since to fetch only what's new since your last run.

Reviewed signals showing your draft vs. the final human-edited/published text, plus attached feedback. The primary 'learn from outcomes' source.

Search saved learnings (pinned first). Use to recall durable knowledge before drafting.

List captured signals/leads, newest first. Use to answer 'what came in', 'what's pending review', or to inspect specific leads.

List ingested briefs (full raw markdown), newest first.

Platform analytics summary: funnel, per-community/tier/platform breakdowns, strong buys, outcomes, feedback counts, response-time medians.

Persist a durable lesson the platform should remember (feeds future context packs). Use for insights worth carrying across runs — not per-signal notes.

Submit one buy-signal/lead directly (deduped by external_url). Provide draft replies as variants. confidence >= 0.8 alerts the team immediately. Prefer one call per signal over writing brief files.

Move a signal to a new review status (multi-hop paths like new->approved are resolved server-side). Use for operator-requested triage, e.g. archiving stale signals. Look up by id or external_url.

Submit a whole brief as raw markdown (alternative to signal_submit / writing a vault file). Prefer signal_submit for structured leads.