curate

Curate — Judge Pending Offers Against the Host Vault

An offer (a served engram learn from a child, or a parent note pulled down by engram activate) lands as a pending offer: a real note file, marked pending: true, that hasn't yet been reviewed against the vault's existing (non-pending) notes. Curation is that review — the same agent-judged covered/near/absent reasoning recall's Step 2.5 performs against query candidates, applied instead to a pending offer against the host vault's existing notes.

Host-local only, on your own initiative. engram serve's only job on an offer is stamp the declared identity, persist the pending marker (or update the same origin's pending offer in place), respond; a pull-down's only job is fetch and persist the pending copy — judgment happens later, off the request path, never synchronously inside the HTTP request that created the offer. Invoke this skill reactively, whenever you notice one of the three surfacing signals (query payload flag, update notice, write-path nudge), or whenever explicitly asked to curate.

Special Case — Pending Skill Runbook Notes

A pending runbook note that carries a skill_hash field is a skill's registration offer, not a served write. Handle it separately from regular pending offers — never judge it covered/near/absent, never discard it, and always keep it in the vault.

For new notes (no runbook fields yet): Author its situation, done_when, triggers, and red_flags from the skill's body:

• situation: when would an agent reach for this skill (retrieval-shaped phrasing, e.g. "reviewing code for style violations")
• done_when: the observable end state (e.g. "all flagged issues fixed or documented")
• triggers: literal user phrasings that should fire it (e.g. "code review", "style check"; may be empty)
• red_flags: the skill's key failure modes — aim for the 1200-byte engram query preview budget (rendered YAML bytes); never drop a valid red flag to fit

Amend the note: engram amend --target <basename> --situation="<text>" --done-when="<text>" --red-flag="<text>" --trigger="<phrase>" (repeat --trigger and --red-flag for multiple entries). A free-text flag value (for example --situation, --source, --subject, --behavior) must be written as --flag="value" — the quote touching the =, no space — because a value beginning with - is otherwise misread as a new flag and the command fails. Then clear the marker: engram amend --target <basename> --clear-pending (a skill note has no xid, so engram show prints no # exchange_hash line and there is no --expect-hash to pass).