Knowledge Packs
Packs for remembering, retrieving, and researching — MemPalace, Telos, Research, ContextSearch.
Four Packs cover the knowledge surface. MemPalace remembers what was said. Telos tracks who you are and what your projects depend on. Research goes out and finds what you don't know. ContextSearch finds what you already did.
MemPalace
What it is — a local-first semantic memory built on the Method of Loci. Verbatim conversation content lives inside a Wing → Hall → Room → Closet → Drawer hierarchy, with cross-wing tunnels wiring related rooms together.
When to use it — recall a past decision, mine the knowledge graph for connections, or check what's been filed. Triggers include "what did we decide about pricing three weeks ago," "find every time I talked about the auth migration," and palace status.
Example outcome — you ask "what was the final call on the Stripe checkout flow?" and MemPalace returns the verbatim drawer from the pricing-decision room in wing_myproject, plus tunnels pointing to related drawers in two other wings. No rewriting, no summarization — the exact words from the original session.
Hooks run this Pack automatically. MemPalaceWakeUp injects L0+L1 context at SessionStart. MemPalaceLearn files learnings at SessionEnd. MemoryHarvest extracts corrections and decisions from the transcript. Direct invocation is for search and graph queries.
Telos
What it is — TELOS (Telic Evolution and Life Operating System) runs in two modes. Personal TELOS tracks beliefs, goals, lessons, books, and movies at ~/.claude/DOS/USER/TELOS/. Project TELOS analyzes a directory and produces dependency maps, entity extraction, and McKinsey-style executive reports.
When you'd reach for it
You want to add a book to your reading log, map the dependencies of a new consulting engagement, or produce an n=24 narrative-points report from last week's stakeholder interviews. Telos has a workflow for each.
Example outcome — you run Skill("Telos", "analyze ~/Projects/MyApp") and Telos scans every .md and .csv in the tree, extracts entities and relationships, produces a progress dashboard, and writes a timestamped McKinsey report into the project directory. Backup of any modified file lands alongside.
Research
What it is — a multi-agent research engine with four depth modes. Quick runs 1 agent, Standard runs 3, Extensive runs 12 in parallel across Claude, Gemini, Grok, and Perplexity, and Deep Investigation iterates across multiple vault builds. Backed by 242+ Fabric analysis patterns and a mandatory URL verification protocol — every link is checked before delivery because research agents hallucinate URLs.
When to use it — any time you say "research." The word itself is the trigger. "research" alone routes to Standard. "quick research on X" routes to Quick. "extensive research on Y" fires the full 12-agent fleet. "extract wisdom" and "use fabric" route through the same Pack.
Example outcome — you ask for extensive research on a new vertical and get a structured brief: 12 agents' findings synthesized against Fabric analysis patterns, every URL verified, confidence tags on each claim, and a source register at the bottom. The whole run is reproducible from the saved vault.
ContextSearch
What it is — a unified search across prior DOS work. One query hits session registry, work directories, PRDs, git history, session names, and Claude Code conversation history in one pass. Runs as /cs or /context-search.
When you'd reach for it
You half-remember solving this exact problem three weeks ago in a different session. Run /cs rate limiting and ContextSearch pulls the three sessions that touched it, loads the PRD, and waits for your next instruction with the context already primed.
Example outcome — you run /cs authentication and get a ranked list of past sessions that touched auth, the most relevant PRD loaded into the current context, and a prompt to continue. Pair it with a task — /cs authentication followed by "now add rate limiting to the gateway" — and the next request runs with full prior context already resolved.
Was this page helpful?