toolhutch

name	toolhutch
description	Audit local agent tool manifests and explain capability risk before enabling tools.
version	0.1.0
targets	["codex","claude","openclaw","agents"]
tags	["security","tools","mcp","permissions","review"]

toolhutch

Use this skill when an agent or maintainer needs a local evidence brief for MCP, OpenClaw, or JSON/YAML tool manifests before enabling tool access.

Inputs

• A local file or directory containing JSON or YAML tool/server definitions.
• An optional local policy file with allow, warn, or deny rules.
• A requested output format: Markdown for review, JSON for automation.

Tools

• toolhutch scan <path> emits a Markdown risk brief.
• toolhutch scan <path> --format json emits deterministic JSON.
• toolhutch scan <path> --fail-on high exits 2 when the highest detected risk reaches the threshold.
• toolhutch explain <path> --json includes low-risk evidence and exits 2 for critical capabilities.
• toolhutch policy <path> --policy <policy.json> applies local rules and exits 3 on deny.

Side Effects

All commands are read-only. The CLI does not make network calls, execute listed tools, read secret values intentionally, enforce runtime permissions, or upload reports.

Approval Boundaries

Ask for explicit approval before enabling a scanned tool server, sharing a report that might contain sensitive path names, or using the result as a hard security approval. Treat the output as review support, not proof of safety.

Validation

Run:

npm run release:check

For a fixture smoke:

toolhutch scan ./fixtures/risky-openclaw-tools.json
toolhutch scan ./fixtures/risky-openclaw-tools.json --fail-on high
toolhutch policy ./fixtures/risky-openclaw-tools.json --policy ./examples/toolhutch.policy.json

Example

toolhutch scan . --format markdown --policy ./toolhutch.policy.json
toolhutch explain ./mcp.json --json