agentvet-action

Lint LLM tool definitions in your repo. Catches missing descriptions, non-snake_case names, and missing input schemas before the model ever sees a broken tool. Wraps the npm-published @mukundakatta/agentvet library.

Supports Anthropic (input_schema), OpenAI (parameters), and MCP (inputSchema) tool shapes, plus MCP server configs (mcpServers.<name>.tools[]).

Quick start

- uses: actions/checkout@v4
- uses: MukundaKatta/agentvet-action@v1
  with:
    tools-glob: "**/tools/*.json,**/mcp.json"

Inputs

Input	Default	Description
tools-glob	**/tools/*.json,**/mcp.json,**/.mcp.json	Comma- or newline-separated globs of JSON files to scan.
fail-on	error	error fails on hard errors only; warning fails on either; none never fails the build.
report-path	agentvet-report.json	Where to write the JSON report.
node-version	20	Node version for the runner.

Outputs

Output	Description
total-tools	Number of tool definitions scanned.
errors	Count of tools with hard errors.
warnings	Count of tools with warnings.
report-path	Path to the JSON report.

Issue codes

Code	Severity	Meaning
E000	error	file is not valid JSON
E001	error	tool is missing name or description
E002	error	name is not snake_case
W001	warning	description shorter than 10 chars
W002	warning	no inputSchema / input_schema / parameters

Example: gate a PR on tool quality

name: Lint LLM tools
on: pull_request
jobs:
  agentvet:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: MukundaKatta/agentvet-action@v1
        with:
          tools-glob: "tools/**/*.json,mcp.json"
          fail-on: error

Example: warning-only mode (just report, never fail)

- uses: MukundaKatta/agentvet-action@v1
  with:
    fail-on: none
- run: cat agentvet-report.json

Sibling actions

Part of the @mukundakatta agent stack:

• agentsnap-action — snapshot tests for tool-call traces
• mcp-stack-validate-action — single CI gate that runs the whole stack

License

MIT