Menu
Consolidate and clean up existing brain memory. Run this periodically (like a sleep cycle) to dedupe, decay, merge, and relocate memories to vault.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Remove old pi session files to reclaim disk space
Testing-focused coding workflow. Extends code-assist with enhanced test generation, coverage analysis, and test quality gates.
Systematic debugging workflow for diagnosing and fixing issues. Follows a structured approach to isolate problems, form hypotheses, and verify fixes.
Generate documentation from code or scratch - READMEs, API docs, changelogs, and user guides. Complements codebase-summary which analyzes existing code.
Extract durable learnings and preferences from the current conversation for memory capture. Use this after a conversation to capture what should be remembered.
Track skill usage with correctness scoring, generate reports, and identify unused or misused skills for pruning.
| name | cleanup-memory |
| description | Consolidate and clean up existing brain memory. Run this periodically (like a sleep cycle) to dedupe, decay, merge, and relocate memories to vault. |
| kind | sop |
Run a "brain sleep cycle" on your existing memories:
When to use:
extract-memoryDifference from extract-memory:
extract-memory = "What should I remember from this conversation?" (captures new)cleanup-memory = "Clean up my existing memories" (organizes old)~/.rho/brain/brain.jsonl)true)last_consolidation) — last_consolidation | <ISO timestamp> | <duration>1)~/.pi/agent/sessions/)10)high)memory_consolidate.last_consolidated_at)List memory by type and count totals:
Determine the lower bound timestamp for session mining.
Extract durable learnings/preferences from user messages in matched sessions.
Confidence policy:
mediumRun brain action=decay to age out old entries.
Identify duplicates/superseded/stale entries and merge candidates.
Importance levels:
Apply the 30-day test: "Would this change a decision I make 30 days from now?"
Auto-remove categories:
Move reference-heavy entries to vault (useful knowledge that doesn't need injection every turn):
Set/update checkpoint timestamp after successful completion.
Report counts before/after, mining window, added/skipped entries, decayed/removed/merged/relocated counts, vault notes created.
Run periodically (suggested: daily or weekly):
/skill run cleanup-memory
With custom parameters:
/skill run cleanup-memory since=7d max_new_entries=5