Scan the machine for installed AI coding CLIs (claude, codex, gemini, copilot, opencode, ollama), detect whether each account is actually active and which models are usable, and cache the result to ~/.claude/model-inventory.json with role→model routing chains (planner/coder/scout/reviewer/fixer) that goal-runner and autopilot consume to pick the best available model per sub-agent while staying cost-efficient. Three evidence tiers — installed (binary found), likely-authenticated (zero-token credential heuristics), verified (a live one-line probe per model, the only ground truth for "account active and model on the plan"). Use this skill whenever the user says "scan available models", "which models can I use", "what AI CLIs are installed", "refresh the model inventory", "is fable available", "check my model access", or "/model-inventory" — even if they don't name the skill. Not for picking a model mid-task (consumers read the cached file, they don't rescan).
Integrations
ModelBound
Try itPull team skills, rules, and knowledge from the ModelBound hosted MCP server. Use whenever the user references their ModelBound team, asks to sync skills/rul...
What it does
Pull team skills, rules, and knowledge from the ModelBound hosted MCP server. Use whenever the user references their ModelBound team, asks to sync skills/rules, search team knowledge, or propose edits to a team-managed `SKILL.md` / `AGENTS.md` / `CLAUDE.md`.
The skill document
ModelBound
ModelBound is the source of truth for AI team skills, rules, system prompts, and knowledge bases. This skill connects OpenClaw to ModelBound's hosted MCP server so you can search and pull team context on demand instead of copy-pasting files.
Setup
Required env:
MODELBOUND_API_KEY— get one at
Optional:
MODELBOUND_TEAM_ID— scope all calls to a specific teamMODELBOUND_ENDPOINT— defaults tohttps://mcp.modelbound.co
The MCP server is plain HTTPS JSON-RPC. No daemon, no shell-out.
Lean-first protocol
ModelBound exposes 45+ tools. Do not advertise them all. Instead, call this small surface and let the server route:
| Tool | Purpose |
|---|---|
modelbound.listTools | List every available tool with its scope. Always call first if you don't know what's available. |
modelbound.callTool | Invoke any tool by name. Use this for everything else. |
help | Plain-English description of a tool or workflow. |
search.all | Hybrid search across skills, rules, prompts, and corpora. |
search.summary | Reranked top-N answer to a question. |
files.get | Fetch a specific file by ID or slug. |
skills.proposeDraft | Propose an edit to a team-managed skill. Returns a review_url. |
gateway.setWorkspace | Scope subsequent calls to one team / repo. |
If a workflow below references a tool not on this list, route it through modelbound.callTool with the name from reference/tools.md.
Workflows
1. Pull a team skill into the working directory
gateway.setWorkspacewith the user's team or repo slug.search.allfor the skill name (orskills.listviamodelbound.callToolfor an exhaustive list).files.getto fetch theSKILL.mdbody.- Write it to
./.claude/skills//SKILL.md(or whatever location the user's IDE expects).
2. Answer a question from team knowledge
search.summarywith the user's question.- If the summary cites file IDs, follow up with
files.getfor the full text. - Cite the ModelBound URL the server returns — never fabricate one.
3. Read a specific file the user mentioned
files.get with the slug or URL fragment. If ambiguous, search.all first to disambiguate.
4. Propose an edit back to ModelBound
Never call skills.update directly. Always use skills.proposeDraft:
files.getthe current version.- Compute the new content locally.
skills.proposeDraftwith{ skill_id, content, summary }.- Show the user the
review_urlit returns. A teammate approves the change in the ModelBound UI.
5. Discover more capabilities
If the user asks for something this skill doesn't cover (evals, agents, corpora ingestion, MCP gateway), call modelbound.listTools and pick from the result. reference/tools.md has a categorized index.
Auth fallback
If a call returns 401 unauthorized, tell the user:
Set
MODELBOUND_API_KEYto a key from , then re-run the request.
Don't try to recover by guessing or by hitting other endpoints.
Safety rules
- HTTP only. Never shell out, never write to ModelBound files on disk other than under the user's IDE skill directory.
- Read tools (
search.*,files.get,*.list) are safe to call freely. - Write tools (
skills.proposeDraft,corpus.upload,webhooks.create, etc.) require explicit user confirmation in chat first. - The skill bans
skills.updateoutright — propose drafts instead.
More
reference/tools.md— categorized index of all 45+ ModelBound MCP tools.reference/examples.md— three end-to-end JSON-RPC recipes.
Related skills
Double-blind comparison of AI model responses — query models in parallel, judge anonymized outputs, rank on merit. Trigger with "mdls" or "modelshow".
Use APIDot model catalog guidance to choose currently available APIDot image, video, chat, music, and 3D model pages, avoid offline model routes, prevent dup...
Allow any combination of agents, humans, or software to approve MCP tool calls that you flag. Prevent your agent from deleting your production database or violating compliance. Use this skill for agents that PROPOSE calls, not agents that approve calls.
Allow any combination of agents, humans, or software to approve MCP tool calls that you flag. Prevent your agent from deleting your production database or violating compliance. Use this skill for agents that APPROVE calls, not agents that propose calls.
Use when preserving, searching, reviewing, or exporting user-owned agent memory across OpenClaw, Codex, Claude, OpenCode, Hermes, Qoder, Obsidian, and Git. Inputs are local memory/chat stores and handoff summaries; outputs are Obsidian sources, memory indexes, context/profile packs, skill inventorie