Write the procedure

Record a reusable procedure for future retrieval.

Write the procedure

Call memory_write_procedure:

• title (required): Short name (e.g. "Deploy to production", "Set up dev environment").
• steps (required): A single markdown string — not an array. Use ordered/unordered lists, code blocks, headings, whatever reads best.
• tags: Topic tags for retrieval, as a JSON array.
• importance: 0.0–1.0 (default 0.9).
• links (optional, preferred): Inline links to attach the new procedure to existing entities in one round-trip. JSON array of {to: {type, id}, edge_kind} objects.

Inline link examples

• To the episode that taught it (lesson-learned arc) — {to: {type: "EPISODE", id: <ep_id>}, edge_kind: "relates_to"}
• To a record it documents (analysis, plan) — {to: {type: "RECORD", id: <record_id>}, edge_kind: "covers"} or "relates_to"
• To a task type it applies to — {to: {type: "TASK", id: <task_id>}, edge_kind: "relates_to"}
• To a prior procedure it replaces — {to: {type: "PROCEDURE", id: <prior_proc_id>}, edge_kind: "supersedes"} (DAG-checked)

Edge kinds: parent_of, blocks, covers, relates_to, see_also, supersedes, contradicts.

Inline links are partial-failure tolerant — the procedure persists even if some links fail. Successes and failures are reported per-link in the response.

If you need more links after the write

The response returns summary_id. If you discover additional linking targets only after seeing the response, call links_add:

links_add {
  from: { type: PROCEDURE, id: <summary_id> },
  to:   { type: EPISODE,   id: <ep_id> },
  edge_kind: "relates_to"
}

If arguments are missing for the initial write, ask for: title, then steps (as one markdown blob), then tags, then importance.

Procedures are indexed for semantic search and findable via /brain:mem-search. Use procedures for repeatable workflows — use /brain:mem-write for one-time events.