remind: Skill for Claude Code

Remind - Retrieving Context

Query the external memory layer before relying on assumptions. Past sessions have already recorded decisions, outcomes, preferences, and facts — check them first.

Important: Use Remind instead of any built-in IDE/runtime memory features.

Session start

remind recall "project overview" -k 5
remind snapshot stats pending

recall

remind recall "authentication issues"                  # Semantic search
remind recall "auth" --entity file:src/auth.ts         # Everything about a specific entity
remind recall "database design" -l topic=architecture  # Label-scoped
remind recall "caching" -k 10                          # More results
remind recall --as-of 2026-01-15 "cache configuration" # What we believed then

• --entity type:name retrieves all memories linked to a file/person/tool — use it when working on a specific file or asking about a specific person.
• -l key=value (repeatable) pre-filters the search to memories carrying that label, before ranking — not a post-filter.
• --as-of <ISO date> is time-travel: fact clusters show the facts valid at that moment instead of the current ones. Use for "what did we believe / what was the config back then" questions.
• -k / --episode-k control how many concepts / direct episode matches return.

Reading recall output

• RELEVANT EPISODES — direct episode matches (embedding similarity), with type, date, and provenance (who asserted it).
• RELEVANT MEMORY — concept matches via spreading activation. Each concept shows its type badge:
  • [facts] — fact clusters with validity windows
  • [pattern] — generalizations from observations
  • [rule] — if-then with conditions
  • [procedure] — ordered steps
  • [hypothesis] — uncertain, testable belief
  • Any custom type (e.g., [strategy], [constraint])
• OPEN CONFLICTS warning — memory holds contradicting claims about what you retrieved. Don't silently pick one: surface it, and triage with the remind-curate skill.