How do I install Agent Readiness?

Copy the stdio config for Cursor, Claude Desktop, Claude Code, or VS Code on this page. It uses the publisher's install, not a PlugMCP proxy.

Does Agent Readiness need an API key?

No required environment variable is listed on this record.

What transport does Agent Readiness use?

Agent Readiness runs locally over stdio.

When was Agent Readiness last checked?

Last checked 2026-11-09. Deterministic, read-only audits of repo readiness for AI coding agents and of agent context files.

Agent Readiness MCP server

agent-readiness-action

What it is

A GitHub Action that checks how ready a repository is for AI coding agents (Claude Code, Cursor, Codex, Copilot, and similar). It bundles two deterministic engines:

agent-readiness-kit scores the repository from 0 to 100: instruction files, architecture notes, scripts, tests, safety boundaries, and more. Lockfiles, version pins, test runners, linters, and formatters are scored on the files of each detected ecosystem (Node.js, Python, Go, Rust). It prints a category breakdown in the log and job summary.
agent-context-doctor, with context-audit: 'true', checks the instruction files themselves: placeholders, risky directives, contradictions, stale commands, pasted secrets, hidden Unicode, and risky Claude Code and MCP configuration such as bypassPermissions, hooks that run remote scripts, or third-party API endpoints.

It can fail the build below a score, block pull requests that lower the score, write a Markdown report, and post a summary comment on pull requests.

Both engines are bundled into the action at pinned commits, so nothing is downloaded or executed from a package registry at run time. There are no external API calls (other than the optional PR comment), no telemetry, and no LLM calls.

Why use it

Catch missing or low-quality agent context (instruction files, architecture notes, setup docs) before it slows down day-to-day agent work.
Catch instruction files and agent settings that steer agents toward unsafe changes or let them run commands without asking.
Track the readiness score over time and gate pull requests on a minimum score or on any drop from the base branch.
Give reviewers a concise, deterministic summary on each PR — a complement to human review, not a replacement for it.

Quick start

# .github/workflows/agent-readiness.yml
name: Agent Readiness Audit

on:
  pull_request:
  push:
    branches: [main]

permissions:
  contents: read

jobs:
  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v7
      - uses: alipajand/agent-readiness-action@v1

Examples

Basic audit

name: Agent Readiness Audit

on:
  pull_request:
  push:
    branches: [main]

permissions:
  contents: read

jobs:
  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v7
      - uses: alipajand/agent-readiness-action@v1

Fail below a minimum score

- uses: alipajand/agent-readiness-action@v1
  with:
    min-score: '70'
    fail-on-threshold: 'true'

Comment on pull requests

permissions:
  contents: read
  pull-requests: write
  issues: write

steps:
  - uses: actions/checkout@v7
  - uses: alipajand/agent-readiness-action@v1
    with:
      comment-on-pr: 'true'
      min-score: '70'
      github-token: ${{ github.token }}

The action posts a single comment the first time and updates it on subsequent runs, so there are no duplicate comments. It only updates a comment that starts with its hidden marker and was written by a bot account, so pasting the marker into your own comment does not make the action overwrite it. When commenting with a personal access token, set comment-author to that account's login.

Block pull requests that lower the score

on:
  pull_request:

jobs:
  audit:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v7
        with:
          fetch-depth: 0 # the base commit must be available
      - uses: alipajand/agent-readiness-action@v1
        with:
          baseline-ref: ${{ github.event.pull_request.base.sha }}
          max-score-drop: '0'

The action audits the base commit in a temporary git worktree, reports the change (▼ -5 vs <sha>) in the log, job summary, and PR comment, and fails when the score drops by more than max-score-drop.

Also check the agent instruction files

- uses: alipajand/agent-readiness-action@v1
  with:
    context-audit: 'true' # run the bundled agent-context-doctor
    context-fail-on: 'high' # fail on risky instructions, secrets, hidden characters, ...

With context-audit, the action also runs agent-context-doctor on the same path. Its score and top issues appear in the log, job summary, and PR comment. It honors the audited repository's .acdrc.

Write a Markdown report artifact

steps:
  - uses: actions/checkout@v7

  - uses: alipajand/agent-readiness-action@v1
    with:
      output: 'docs/agent-readiness-report.md'

  - uses: actions/upload-artifact@v7
    with:
      name: agent-readiness-report
      path: docs/agent-readiness-report.md

Audit a subdirectory or monorepo package

- uses: alipajand/agent-readiness-action@v1
  with:
    repo-path: './packages/web'

Inputs

Input	Default	Description
`repo-path`	`.`	Path to the repository or subdirectory to audit. Relative paths are resolved from the GitHub Actions workspace root.
`min-score`	`0`	Minimum acceptable score (0–100). Checked when `fail-on-threshold` is `true`.
`output`	`''`	Write a Markdown report to this path. Relative paths are resolved under `repo-path`, and the result must stay inside `repo-path`. The report is never written through a symlink.
`json`	`false`	Echo the raw JSON audit output to the Actions log.
`comment-on-pr`	`false`	Post or update a PR comment. Only runs on `pull_request` events. Requires a token with `pull-requests: write` and `issues: write`.
`github-token`	`''`	Token for the PR comment, usually `${{ github.token }}`. Falls back to the `GITHUB_TOKEN` environment variable.
`comment-author`	`''`	Only update an earlier summary comment written by this login. Defaults to any bot account.
`baseline-ref`	`''`	Git ref to compare against, usually `${{ github.event.pull_request.base.sha }}`. Audited in a temporary worktree; needs `fetch-depth: 0`.
`max-score-drop`	`''`	Fail when the score is more than this many points below the `baseline-ref` score (`0` fails on any drop). Requires `baseline-ref`.
`job-summary`	`true`	Write the audit summary to the workflow run's job summary.
`context-audit`	`false`	Also check agent instruction files with the bundled agent-context-doctor.

Outputs

Output	Description
`score`	Final agent-readiness score (0–100).
`report-path`	Absolute path of the written Markdown report, or empty when `output` was not set.
`passed`	`"true"` when the score is at least `min-score`, otherwise `"false"`.
`categories`	JSON array of category scores: `id`, `label`, `score`, `maxScore`.
`baseline-score`	Score at `baseline-ref` (only when `baseline-ref` is set).
`score-delta`	Score minus the baseline score (only when `baseline-ref` is set).
`context-score`	agent-context-doctor score, when `context-audit` is `true`.
`context-grade`	agent-context-doctor grade: `excellent`, `good`, `needs-work`, or `risky`.
`context-issue-count`	Number of agent-context-doctor issues.

Permissions

Basic use (audit, log, and threshold check) only needs read access:

permissions:
  contents: read

To post PR comments with comment-on-pr: "true", the job additionally needs pull-requests: write and issues: write, and the step must pass a token through the github-token input (or the GITHUB_TOKEN environment variable):

permissions:
  contents: read
  pull-requests: write
  issues: write

PR comments are created and updated through GitHub's issue-comment API (issues.listComments / issues.createComment / issues.updateComment), which is why issues: write is included alongside pull-requests: write.

How it works

Runs the bundled agent-readiness-kit audit engine against repo-path in the runner. The engine is compiled into dist/index.js from the vendor/agent-readiness-kit submodule, pinned to a reviewed commit.
Collects the result (score, categories, findings, missing items, recommendations).
Logs a summary and a collapsible detail group.
If output is set, writes the kit's Markdown report inside repo-path.
If json is true, echoes the raw JSON to the log.
If baseline-ref is set, audits that commit in a temporary git worktree with the same bundled engine and computes the score change.
If comment-on-pr is true and the event is a pull_request, posts or updates a comment.
Writes the summary to the job summary (unless job-summary is false).
Marks the step as failed if score < min-score (with fail-on-threshold) or the score dropped by more than max-score-drop.

Security

The action is meant to run on pull requests, including ones from contributors you don't fully trust. Both engines are bundled from pinned submodules (vendor/agent-readiness-kit, vendor/agent-context-doctor):

No runtime downloads. Earlier versions ran npx --yes agent-readiness-kit, which fetched whatever package owned that name on npm. That name belongs to an unrelated project. The audit engine is now bundled from a pinned commit of agent-readiness-kit.
Contained writes. The Markdown report must resolve inside repo-path and is never written through a symlink. The audit does not write score history into the repository.
Log and comment safety. Audit details are logged with workflow commands paused, so a crafted file name cannot emit ::error:: or other commands. PR comments escape HTML and keep file names in inline code.
Comment ownership. Only a comment that starts with the action's marker and was written by a bot account (or comment-author) is updated.

See SECURITY.md to report a vulnerability, and docs/ARCHITECTURE.md for the module layout and trust boundaries.

Release checklist

A release has two separate steps: publishing an immutable version tag, and promoting it through the floating major tag. Workflows that use @v1 change only at the second step.

Publish the version:

Run git submodule update --init so both engines under vendor/ are present.
Run pnpm format:check, pnpm test, and pnpm typecheck.
Run pnpm build.
Commit the updated dist/ if it changed. CI fails when dist/ does not match the source.
Set version in package.json and move the Unreleased notes in CHANGELOG.md under the new version.
Tag the release commit on main with its full version (for example v1.2.0) and push only that tag: git tag -a v1.2.0 -m "agent-readiness-action 1.2.0", then git push origin v1.2.0. Never move or re-create a version tag.

Promote it to v1 only after it has passed a real consumer workflow:

In a repository that uses the action, pin the new version's commit SHA on a branch and open a draft pull request, keeping the workflow's inputs, permissions, and baseline-ref. Check that the pull request run and the baseline comparison pass, and compare the scores, log, annotations, and context findings with the previous version.
Move the major tag to the tested commit: git tag -f v1 v1.2.0^{}, then git push -f origin v1.
Check that v1 resolves to that commit: git ls-remote origin refs/tags/v1 must print the commit of v1.2.0^{}.

Workflows that use @v1 pick up a release at step 8; pin a full commit SHA to review every change first. The integrations (skill, plugin, MCP server) are versioned and tagged separately as integrations-v<version> (docs/RELEASE-CHECKLIST.md); those tags never move or depend on the action's v tags.

Outside GitHub Actions

The same pinned engines are packaged for local use by AI coding agents. The scanner is read-only and offline; the agent that runs it may send the results to its model provider (security model):

Agent Skill agent-readiness-audit for Claude Code, Cursor, Codex, and other skill clients: docs/integrations/SKILL-USAGE.md
Claude Code and Cursor plugin agent-readiness: docs/integrations/PLUGINS.md
Local MCP server @alipajand/agent-readiness-mcp: integrations/mcp-package/README.md; listed on MCPRush

See docs/distribution/README.md for what is published where.

Related tools

agent-readiness-kit — CLI used by this action.
agent-context-doctor — checks whether agent instruction files are specific, safe, and usable.
agent-pr-reviewer-lite — flags risky PR diffs before merge.

Development

git submodule update --init   # fetch the pinned engine sources
pnpm install
pnpm format      # prettier (format:check in CI)
pnpm test        # vitest unit tests
pnpm typecheck   # tsc --noEmit
pnpm build       # ncc bundle → dist/index.js

The built dist/ must be committed alongside source changes.

To move to a newer engine, check out the commit you want in vendor/agent-readiness-kit or vendor/agent-context-doctor, run pnpm build, and commit both the submodule pointer and dist/. Dependabot opens weekly submodule update PRs; those need a pnpm build commit before CI passes.

See CONTRIBUTING.md for the full workflow and pull request guidelines.

License

MIT

Agent Readiness

Add this server

Host a copy with our keys and logs.

Server facts

Tools

Questions