memory

Interact with the project's cognitive memory system (shodh-memory).

Usage: /set:memory [subcommand] [args]

Subcommands:

Core:

• status [--json] — Show config, health, and memory count
• recall <query> [--limit N] [--mode MODE] [--tags t1,t2] [--tags-only] [--min-importance F] — Semantic search
• proactive <context> [--limit N] — Context-aware retrieval with relevance scores
• remember --type TYPE [--tags t1,t2] [--metadata JSON] [--failure] [--anomaly] — Save a new memory (stdin)
• list [--type TYPE] [--limit N] — List all memories (JSON)
• browse (default) — Show summary grouped by type
• projects — List all projects with memory counts

Forget / Cleanup:

• forget <id> — Delete a single memory by ID
• forget --all --confirm — Delete ALL memories
• forget --older-than <days> — Delete memories older than N days
• forget --tags <t1,t2> — Delete memories matching tags
• forget --pattern <regex> — Delete memories matching regex

Diagnostics:

• stats [--json] — Memory quality diagnostics (types, tags, importance)
• cleanup [--threshold F] [--dry-run] — Remove low-value memories (default threshold 0.2)
• audit [--threshold F] [--json] — Duplicate detection report (default threshold 0.75)
• dedup [--threshold F] [--dry-run] [--interactive] — Remove duplicate memories

Introspection:

• get <id> — Get a single memory by ID (JSON)
• context [topic] — Condensed summary by category
• brain — 3-tier memory visualization

Export / Import:

• export [--output FILE] — Export all memories to JSON
• import FILE [--dry-run] — Import memories from JSON (skip duplicates)

Sync (team sharing via git):

• sync — Push + pull in one step
• sync push — Export and push to git remote
• sync pull [--from user/machine] — Pull and import from git remote
• sync status — Show sync state and remote sources

Maintenance:

• health — Check if shodh-memory is available
• health --index — Check index health (JSON)
• repair — Repair index integrity
• migrate — Run pending memory migrations
• migrate --status — Show migration history

What to do:

1. Parse $ARGUMENTS to determine the subcommand (first word). If no arguments, default to browse. If unrecognized subcommand, pass through to set-memory directly.

2. Check health first:

   set-memory health
   

   If this fails (exit code non-zero), display:

   "Memory system not available — shodh-memory is not running. Check with set-memory health." Then STOP.

3. Execute the subcommand:

   status:

   set-memory status $REMAINING_ARGS
   

   Display the output directly.

   recall (rest of arguments is the query and flags):

   Recall modes (--mode): semantic (default), temporal, hybrid, causal, associative. Parse the JSON output. For each memory, display:

   • Type and tags
   • Content (truncated to 200 chars if long)
   • Creation date If no results, say "No matching memories found."

   proactive (rest of arguments is context string):

   set-memory proactive "<context>" --limit 5
   

   Parse JSON and display results like recall.

   remember (rest of arguments is the content):

   • Use the AskUserQuestion tool to ask which memory type to use:
     • Options: Decision, Learning, Context
   • Optionally ask for comma-separated tags (or skip)
   • Then run:

   • Confirm: "Memory saved."

   list:

   set-memory list $REMAINING_ARGS
   

   Parse JSON and display in a readable table format.

   browse (default):

   set-memory status
   set-memory list
   

   Display status summary, then group memories by type (Decision, Learning, Context) and show counts + recent entries.

   forget (pass all remaining args through):

   set-memory forget $REMAINING_ARGS
   

   DESTRUCTIVE — warn user before running without --dry-run. Display output directly.

   stats / cleanup / audit / dedup (pass all remaining args through):

   set-memory <subcommand> $REMAINING_ARGS
   

   Display the output directly. For dedup without --dry-run or --interactive, warn about destructive operation.