Agent Edl
Natural language → a PROPOSED edit of a timeline (#127). Gated on agentic + projects:write.
Proposes only — it never writes. The caller applies the returned edl by PUTting it, so
validate_edl + _reject_foreign_assets stay the only path an EDL takes into the database.
Returns the NORMALIZED proposal (what the whitelist kept) plus a per-op report saying what
applied and what didn’t — the same honesty contract as the PUT’s normalized echo.
projects:write rather than projects:read: it writes nothing, but its entire purpose is to
produce a document to be written, and it costs a model call. A read-only key should not be able
to spend one.
Headers
Path Parameters
Body
#127 twin. edl is the timeline to plan against — required, and NOT read from the DB: a
programmatic caller may be composing a document it has not saved (and the cookie route plans
against the editor's unsaved on-screen EDL for the same reason). ops re-applies a plan the
caller already has, skipping the model call.
Response
Successful Response
The response is of type Response Agent Edl Api V1 V1 Projects Public Id Edl Agent Post · object.