remember

name	remember
description	Save an explicit scoped memory through Signet. Use only when the user clearly asks to remember something or when a tool/harness must persist an explicit memory row.
user_invocable	true
arg_hint	[critical:] [[tag1,tag2]:] content to remember
builtin	true

/remember

Use this skill for explicit memory writes. Do not use it as a general extraction, ontology, or dreaming path.

Signet's current memory model is source-backed and scoped. A direct remember call creates or deduplicates a memory row, records provenance fields when provided, embeds the content when the daemon can, links mechanically recognizable mentions to existing same-agent entities, and enqueues the normal pipeline work. Raw text does not automatically create new ontology entities, aspects, grouped claims, dependencies, or supersession chains.

When To Use

Use /remember when:

• the user explicitly says to remember, save, persist, or store something
• a harness compatibility path needs to write an explicit memory inside a session
• you need a durable recall row with clear tags, hints, scope, or provenance
• you are importing a single explicit fact and can supply source metadata

Do not use /remember when:

• the task is ontology maintenance or dreaming; use the dreaming skill and signet ontology ... commands
• the task is source ingestion; use signet sources ... or the connector import path so provenance remains inspectable
• the content is a vague behavioral lesson; propose an AGENTS.md, identity, or skill patch instead
• the user did not ask for persistence and normal session capture is enough

CLI

The common path is the CLI wrapper:

signet remember "<content>"

Useful options:

signet remember "<content>" --agent codex --private
signet remember "<content>" --tags signet,recall --hint "future search cue"
signet remember "<content>" --importance 0.9
signet remember "<content>" --critical

Options:

• --agent <name> sets the owning agentId
• --private sets visibility: "private" instead of global
• --tags <tags> stores comma-separated tags
• --hint <hint> adds prospective recall hints; repeat for multiple hints
• --importance <n> sets importance from 0 to 1
• --critical pins the memory
• --who <who> records who is remembering

The daemon must be running:

signet status
curl -s http://localhost:3850/health

API

For provenance, imports, or structured callers, use the canonical endpoint:

curl -s http://localhost:3850/api/memory/remember \
  -H 'content-type: application/json' \
  -d '{
    "content": "User prefers vim keybindings.",
    "agentId": "codex",
    "visibility": "global",
    "tags": "preference,editor",
    "sourceType": "manual",
    "sourceId": "session-key",
    "sourcePath": "memory/codex/transcripts/session.jsonl",
    "idempotencyKey": "stable-import-key",
    "hints": ["editor preference", "vim keybindings"]
  }'

Important fields:

• agentId scopes ownership; do not hardcode default when a real agent is known
• visibility is global or private
• sourceType, sourceId, sourcePath, runtimePath, and idempotencyKey preserve import/source provenance
• createdAt should reflect source time for older imported records
• structured.entities, structured.aspects, and structured.hints are for callers that intentionally author structured graph-adjacent data

When structured is omitted, Signet stores the memory conservatively. It does not invent new graph structure from raw text.

Response Handling

A successful response includes the memory id and whether the row was embedded or deduplicated:

{
  "id": "uuid",
  "type": "preference",
  "tags": "preference,editor",
  "pinned": false,
  "importance": 0.9,
  "content": "User prefers vim keybindings.",
  "embedded": true,
  "deduped": false
}

Confirm the write plainly:

saved: uuid (embedded)

If deduped: true, say that Signet returned the existing row instead of creating a duplicate.

Hard Rules

• Thread agentId and visibility deliberately.
• Preserve source provenance when the memory came from an artifact, transcript, note, import, or external system.
• Use idempotency keys for repeatable imports.
• Do not use remember as a shortcut around ontology operations, source import, or reviewed skill/identity patches.
• Do not silently persist private, sensitive, or authority-changing content without a clear user request or explicit tool contract.