remind-context

name	remind-context
description	Retrieve context from Remind memory before acting. Use at session start, before answering questions about prior decisions, preferences, or project history, when starting work on a familiar file/area, or when asked "what do we know about X" or "what did we believe at time T".

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 (grouped by topic)
remind recall "auth" --entity file:src/auth.ts         # Everything about a specific entity
remind recall "database design" --topic architecture   # Topic-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.
• --topic restricts to a knowledge area (cross-topic results are penalized, not excluded). Without it, results are grouped by topic.
• --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.

Evidence-weighted ranking: Concepts with more supporting evidence rank higher in recall; concepts with contradicting evidence are penalized. A concept's activation score factors in its net evidence strength.

Semantic collision warnings: When a new fact is remembered with high semantic similarity to existing facts (not just entity overlap), recall may surface both as potential conflicts.

Trust currently-valid facts over your assumptions; they carry provenance and supersession history.

Understanding recall output

RELEVANT EPISODES (similarity score 0.0-1.0)

• Higher score = closer embedding match to your query
• Episodes are raw captures — may contain noise or duplicates
• Use for: recent context, exact quotes, provenance checking
• Score >0.9: very close match; >0.8: related; <0.7: tangential

RELEVANT MEMORY (activation score 0.0-1.0)

• Higher score = more relevant via spreading activation through the concept graph
• Concepts are curated knowledge — higher signal than raw episodes
• Use for: established patterns, decisions, generalizations

Concept type badges

• [facts] — Fact cluster with temporal validity; check valid_from/valid_to
• [pattern] — Generalized observation from multiple episodes
• [rule] — If-then relationship with conditions
• [hypothesis] — Uncertain belief, testable
• Any other badge — Custom concept type

When you see OPEN CONFLICTS

Memory contains contradicting claims about the retrieved topic. Do not silently pick one.

1. Surface the conflict to the user, OR
2. Run the curation workflow to resolve it:
   • remind snapshot conflicts to see details
   • remind apply 'resolve id=<conflict_id> winner=<fact_id> note="reason"'

snapshot (structured reads)

For machine-readable memory state, use snapshot with combinable scopes:

# Core scopes
remind snapshot stats                         # Memory statistics
remind snapshot pending conflicts             # Pending episodes + open conflicts
remind snapshot health                        # Actionable issues summary

# Browsing scopes (for exploring memory)
remind snapshot concepts                      # All concepts
remind snapshot episodes:20                   # Recent 20 episodes
remind snapshot entities                      # All entities with mention counts
remind snapshot entities:person               # Filter entities by type
remind snapshot topics                        # All topics with stats
remind snapshot decisions:10                  # Recent 10 decision episodes
remind snapshot questions                     # Open question episodes

# Detail scopes
remind snapshot entity:concept:caching        # All data for an entity
remind snapshot topic:architecture            # All data for a topic
remind snapshot concept:abc123                # Concept detail with facts, history
remind snapshot "query:authentication issues" # Semantic search as JSON

snapshot returns JSON — ideal for feeding into apply changesets or automated processing.

Browsing (no query yet)

remind snapshot stats                        # Memory size/health overview
remind snapshot health                       # Actionable issues needing attention
remind snapshot concepts                     # List all concepts
remind snapshot entities                     # Entities available for --entity recall

Explore workflow: snapshot stats health → recall "<query>" → snapshot concept:<id> for details.