ai-memory

Detect content drift of an operator's scaffolded sanctum files (BOND, CAPABILITIES, CREED, INDEX, LORE, MEMORY, PERSONA, PULSE) against the evolving reference templates, and surface recommended add/remove WITH rationale — never a silent overwrite. Use on a session-start drift check, after the reference templates change, or when the operator asks whether their sanctum is current.

Check ai-memory system status and collection stats

Check ai-memory system status and collection stats

Manually save current session context to ai-memory

Search ai-memory for relevant stored memories

Search ai-memory for relevant stored memories

Design agent team structure for parallel work execution

Enforce oversight-file caps at session-close, rotate over-cap append-only-logs and registers into dated archive shards, and fix over-cap files non-interactively (including MEMORY.md lossless relocation).

Research current best practices for any technology, pattern, or coding standard. Use when asking about best practices, conventions, coding standards, recommended approaches, or how should I questions. Searches local knowledge first, then web for current sources (prioritizing the last ~6 months relative to today). Evaluates if findings warrant a reusable skill.

Search memory system with advanced filtering and intent detection

Research current best practices for any technology, pattern, or coding standard. Use when asking about best practices, conventions, coding standards, recommended approaches, or how should I questions. Searches local knowledge first, then web for 2024-2026 sources. Evaluates if findings warrant a reusable skill.

Two-phase capped startup context loaders for Parzival (activation + session-start)

Creates and maintains an up-to-date wiki of the user's current project, in-session, while you work — a dedicated wiki/ directory with a quickstart entrypoint plus section pages, grounded in source and git evidence. Claude Code itself authors the pages; a Python engine does the deterministic work (repo inventory, wiki scaffold, content-hash/gitHead state, git-diff-since-last-run, CLAUDE.md/AGENTS.md pointer, verifier manifest). Modes init | update | status, plus verify and finalize. Use when asked to create, refresh, or check the freshness of the project wiki/docs. Do NOT use for AI Memory search/save/health (use aim-search, aim-save, aim-status) or SOT drift (use aim-sot).

Scan oversight tracking files, regenerate bug/TD INDEX views, report STATUS divergences, verify phantom-open bug candidates against git history, and check decision-log body coverage.

Tracks the source-of-truth for each part of the user's own project — registry schema, templates, and engine (consult / detect-propose / verify). Detects file and directory drift (content-digest and directory tree-digest), runs a machine-local shadow-git history for change detection, and correlates code changes to stale docs (doc-drift via DOCOWNERS). Use when checking SOT drift, choosing a drift_strategy, setting up the shadow git, or reviewing doc-staleness findings. Do NOT use for AI Memory search or save operations, or for checking AI Memory system health (use aim-search, aim-save, or aim-status).

Generic and BMAD agent selection, instruction preparation, and activation routing

tmux agent lifecycle management for non-Claude providers

Select the appropriate LLM model for each agent based on task complexity and role

Load Parzival cross-session memory from Qdrant

Enforce hygiene on per-operator sanctum files (LORE/MEMORY compaction, PERSONA Evolution-Log section-cap, CREED/BOND cap-check) — line-cap enforcement, anchored-summarization compaction at ~80% of cap, the prune-vs-archive decision, and a conservation 0-lost proof on every relocation. Use when a sanctum file approaches its size cap, on a scheduled hygiene pass, or when the operator asks to compact or prune lore.

Save a Parzival session decision to Qdrant for cross-session memory and L2 retrieval

Initialize sanctum directory for an agent. Runs deterministic First Breath scaffolding — creates folder structure and writes template-substituted files.

Search GitHub issues, PRs, commits, and CI results with semantic search and filters

Check AI Memory system status

Search GitHub issues, PRs, commits, and CI results with semantic search and filters

Synchronize GitHub issues, PRs, commits, and CI results to AI Memory

Search Jira issues and comments with semantic search and filters

Synchronize Jira issues and comments to AI Memory

Load Parzival behavioral constraints as context reminder

Scan code-patterns collection for stale memories by comparing against GitHub code blob data

Toggle auto_update_enabled kill switch for automatic memory updates

Purge old memories from Qdrant collections with safety guards

Manually re-evaluate freshness for code-patterns memories

Manually save current session to memory

Display current memory system configuration and settings

Save Parzival session handoff document to Qdrant for cross-session memory

Save a Parzival insight or learning to Qdrant for cross-session memory

Generic agent instruction preparation and activation -- Layer 3a

Shared agent lifecycle management -- send, monitor, review, accept/loop, shutdown, summary

Select the appropriate LLM model for each agent based on task complexity and role