This file is the Master Registry of all high-performance tools, integrations, and specialised skills available to your AI assistant. It makes the invisible capabilities of the system visible, ensuring the AI can leverage its full range.
⚡ Integrated CLIs & Environments
The AI has access to the following command-line tools or client execution APIs in this environment:
1. Terminal / Shell Access
- Strategic Intent: File operations, folder structures, and script execution — granted via the Desktop Commander connector on Claude Desktop. The user never touches a terminal; the AI runs commands with approval.
- Status: [Enabled / Disabled / Read-Only]
2. Web & Research Fetch
- Strategic Intent: Connecting to APIs, downloading raw markdown documentation, and scraping research links.
- Status: [Enabled / Disabled]
🧩 Active Skills (Procedural Workflows)
Specialised multi-step procedures stored in /SKILLS. The AI reads the corresponding skill file before starting any of these tasks:
- System Audit (sys-audit.md): Diagnostic review of current knowledge storage and decision-making bottlenecks.
- Decision Log (decision-log.md): Framework for recording context, options, and rationale for major decisions.
- AI OS Setup (vault-setup.md): Guided bootstrap of a personal AI operating system vault (
<name>-ai-os) — interview, blueprint, inside-out build, plus anti-rot maintenance. - Tool Manager (tool-manager.md): Structured assistant workflow to look up, install, configure, and record new tools.
- Session Protocol (session.md): Open and close working sessions — orient from memory at the start, persist changes to
memory.mdat the end.
Changelog
2026-08-22 — CLI Policy Clarified
- Added “CLI Requirements” to
tool-manager.md: zero CLIs required by default (Desktop Commander uses commands built into macOS/Windows); git is the only earned exception, with agent-driven install paths verified Aug 2026 (macOS: first git command triggers Apple’s Command Line Tools dialog; Windows:winget install --id Git.Git). vault-setup.mdStep 5 now offers an optional git “safety net” — version snapshots of the new vault, driven entirely by the agent.
2026-08-22 — Added Session Protocol
- Added
session.md: session-start orientation (bootstrap → user → memory → tools) and session-end persistence (updatememory.md, prune stale entries, bumplast-updated/revisitdates). - Registered in
AGENTS.md§3,index.md,README.md; correctedAGENTS.md§4 to list only the prompts that actually ship.
2026-08-22 — Replaced Weekly Review with AI OS Setup
- Removed
weekly-review.md(generic end-of-week ritual with no structural role). - Added
vault-setup.md: bootstraps a personal<name>-ai-osvault in Documents — five-layer inside-out model (identity → rules → skills → agents → tools on a Wiki/memory substrate), interview →os-blueprint.md→ confirm → build or blueprint-only paths, and an anti-rot maintenance cadence (fix-on-contact, weekly drift report, monthly revisit walk). - Updated skill routing in
AGENTS.md,index.md, andREADME.md.
2026-08-22 — Tool Setup Modernised
- Rewrote
tool-manager.mdaround Claude Desktop’s Connector Directory: one-click installs only (Desktop Commander, Filesystem, Gmail/Microsoft 365, Google Drive), no config files for non-technical users. - Updated
AGENTS.md§6 andTOOLS.mdto match (ChatGPT Apps/Developer mode realities, Antigravity MCP Store, terminal commands flagged as high-risk). - Added Codex as the ChatGPT-side path for local file access; noted retirement of ChatGPT Agent mode (replaced by ChatGPT Work / cloud browser).
2026-08-20 — Added Tool Manager Skill
- Added the
tool-manager.mdskill workflow to enable interactive environment audits and tool installation. - Cleaned and generalised
TOOLS.mdandCAPABILITIES.mdto serve as clean starter templates.