Command Claude Code
Plain-text persistent memory system for AI agents. Conventions for L0 progressive loading, three-tier storage (hot/warm/glacier), single-source-of-truth, temporal validity, and wiki-links. Run /cog to bootstrap or reconfigure domains.
Command Claude Code
Create a git commit with Conventional Commits format, selective staging, and secret/artifact guard rails. Invoke with /commit. Repo utility — not part of the core memory pipeline.
Command Claude Code
Audit memory architecture, review rule effectiveness, measure metrics, and propose system improvements. Run monthly to keep the system healthy. Invoke with /evolve.
Command Claude Code
Writing and drafting — systematic clarity (Atkins) fused with writing-as-discovery (Montaigne). For articles, essays, posts, talks, and explanations. Invoke with /explainer. Bundled writing extra — not part of the core memory pipeline.
Command Claude Code
Cross-domain strategic scan. Detects convergence between life areas, velocity changes, and upcoming timing windows. Writes one actionable nudge per run. Invoke with /foresight.
Command Claude Code
Deep memory search and recall — recursive search across all memory files, piecing together a narrative from observations, entities, and action items. Invoke with /history or on "what did I say about..." queries.
Command Claude Code
Archive old observations, prune stale items, rebuild indexes, enforce entity format, and maintain memory health. Run weekly to prevent memory bloat. Invoke with /housekeeping.
Command Claude Code
De-AI text — remove AI writing artifacts and inject human voice, based on Wikipedia's "Signs of AI writing" guide. Edits existing text only. Invoke with /humanizer. Bundled writing extra — not part of the core memory pipeline.
Command Claude Code
Activation shim for the personal memory domain — family, health, calendar, day-to-day logistics. Wakes domain memory when the conversation involves this domain. All routing behavior lives in the cog skill. Generated by /cog — do not hand-edit.
Command Claude Code
Mine recent interactions for patterns and consolidate memory. Detects contradictions, promotes observations to patterns, triages hot-memory, and suggests thread candidates. Run weekly or nightly for best results. Invoke with /reflect.
Command Claude Code
Decision simulation — take a specific decision point, branch it into 2-3 paths, map dependencies, timelines, and canary signals grounded in real memory data. Invoke with /scenario or when foresight flags a candidate.
Settings file Claude Code
Agent settings declaring 19 allowed tools.
Instructions file
Instructions for marciopuga/cog, covering cog — memory system, core conventions, memory path, persona and memory system.