Agent-Skills.md

Agent Skills: Knowledge Stack

Use this skill whenever working inside any of Nick's repos (under ~/Developer/) or whenever a conversation touches Nick's durable cross-cutting work — Atlas Crew product strategy/brand/planning, the Unanet job, job search, interviews, tax amendment, foreclosure/tenant situation, strategic positioning. Maps Nick's three Basic Memory projects (products, work, main) and the per-repo dev trackers, and says which layer owns which kind of question. Consult before reasoning from stored memory summaries about repo architecture, product decisions, active priorities, or ongoing personal/career state — stored memory is lossy; the trackers and the Basic Memory projects are the real artifacts and should be consulted first.

UncategorizedID: NickCrew/claude-cortex/knowledge-stack

Repository

NickCrew/claude-cortex
NickCrewLicense: MIT
2212

Install this agent skill to your local

pnpm dlx add-skill https://github.com/NickCrew/Claude-Cortex/tree/HEAD/skills/knowledge-stack

Skill Files

Browse the full folder contents for knowledge-stack.

Download Skill

Loading file tree…

skills/knowledge-stack/SKILL.md

Skill Metadata

Name
knowledge-stack
Description
"Use this skill whenever working inside any of Nick's repos (under ~/Developer/) or whenever a conversation touches Nick's durable cross-cutting work — Atlas Crew product strategy/brand/planning, the Unanet job, job search, interviews, tax amendment, foreclosure/tenant situation, strategic positioning. Maps Nick's three Basic Memory projects (products, work, main) and the per-repo dev trackers, and says which layer owns which kind of question. Consult before reasoning from stored memory summaries about repo architecture, product decisions, active priorities, or ongoing personal/career state — stored memory is lossy; the trackers and the Basic Memory projects are the real artifacts and should be consulted first."

Knowledge Stack

Nick's durable knowledge lives in two kinds of place: per-repo trackers (code-adjacent) and three Basic Memory projects (everything else). Stored memory summaries are lossy snapshots of these; the artifacts below are the real truth. Consult them before reasoning.

The Layers

Per-repo dev work → the repo's own tracker, not Basic Memory.

  • Architecture, implementation plans, active tasks, roadmap-of-code.
  • Tracker depends on the repo: Cortex uses backlog.md (backlog doc list, backlog task list; see the backlog-md skill). The atlas-crew repos — Facet plus the security repos Apparatus, Chimera, Crucible, Synapse, Bridge — use GitHub Issues + Projects (gh issue, gh project; see the atlas-crew-tasks skill). Don't assume backlog.md everywhere — check the repo.

Atlas Crew product work → the products Basic Memory project (~/Obsidian/Product_Management).

  • Product strategy, brand, positioning, pricing, GTM, planning, specs, PM docs — the product layer, not the code layer.
  • Organized by Atlas Crew family: Security (Synapse, Chimera, Crucible, Bridge, Apparatus), Career Intel (Facet), Dev Tools (Cortex and utilities), Writing (Hard Stuff blog). The root Product Management Vault Home.md is the map.
  • _Archive/ holds frozen history (old job-search material, captured sessions). Treat as read-only context, not current state.

The Unanet job → the work Basic Memory project (~/Obsidian/Work, under unanet/).

  • Personal working-memory for Nick's Senior Platform Release Engineer role: journal/ (append-only daily firehose), systems/, glossary, open-questions, observations-about-me, etc. Its README defines the capture workflow.
  • This project syncs to Nick's work laptop. The bar for what goes here is "fine if a coworker glanced over my shoulder." Candid people-notes, political tripwires, and customer/contract/proprietary specifics stay in segregated local-only files outside this vault — never write them into the synced work project. When in doubt, append to today's journal and route later; when sensitive, keep it out.

Everything else cross-cutting → the main Basic Memory project (~/Obsidian/Default), the default.

  • Personal and career state that isn't an Atlas Crew product and isn't the Unanet job: job search, interviews, tax amendment, foreclosure/tenant situation, strategic positioning, misc research.
  • This is the catch-all. If a topic doesn't clearly belong to products or work, it lands here.

Ephemeral conversation context → this chat. Promote to the right layer when worth keeping.

Picking the project

| Topic | Project | |---|---| | Atlas Crew product strategy / brand / planning / Hard Stuff | products | | Unanet job (people, systems, journal, glossary) | work (syncs to work laptop) | | Job search, interviews, taxes, housing, personal/career | main (default) | | A repo's code architecture, tasks, implementation plans | the repo's tracker (backlog.md / GitHub Issues) |

Operational Rules

  1. Address the right project explicitly. Basic Memory MCP tools and basic-memory tool take a project (e.g. search-notes --project products). main is the default; pass --project work/--project products for the others. Searching the wrong project returns nothing and reads as "no such note."
  2. Before reasoning about a specific repo, list its tracker first (backlog doc list for Cortex, gh issue list for atlas-crew repos) and pull the relevant item before concluding from stored memory.
  3. Before reasoning about cross-cutting work, search the matching Basic Memory project (search, then read the note) before reasoning from stored memory.
  4. When a chat surfaces something durable, propose writing it to the right layer — and use the Basic Memory skills to do it well: memory-notes (frontmatter/observations/relations), memory-ingest (unstructured input → entities), memory-research (research a subject → entity), memory-lifecycle (archive/status moves), memory-schema/memory-metadata-search (structured note types), memory-defrag/memory-reflect (periodic upkeep).
  5. Don't duplicate content across layers. If something could plausibly live in two places (e.g. a product decision that's also a code task), ask Nick which layer owns it.

Failure Modes to Avoid

  • Reasoning from stored memory summaries when fresher truth exists in a tracker or a Basic Memory project — produces stale or wrong answers.
  • Searching one Basic Memory project and concluding the knowledge doesn't exist — it's almost certainly in one of the other two. There are three projects now; check the one that owns the topic.
  • Writing sensitive Unanet material into the synced work project — it reaches the work laptop. Keep candid/proprietary content in the segregated local-only files.
  • Putting Atlas Crew product strategy into a repo tracker, or code-task detail into the products project — knowledge drift across layers that compounds.
  • Treating Obsidian as required — Nick doesn't run the Electron app. The projects are plain markdown; use the basic-memory CLI/MCP or any editor.
  • Enumerating "Nick's repos" or which tracker a repo uses from memory and getting it wrong — ls ~/Developer/ and check the repo (backlog.md vs GitHub Issues).