Agent Done Or Not

Version updated for https://github.com/mohamedzhioua/agent-done-or-not to version v0.13.0.

  • This action is used across all versions by 0 repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

This GitHub Action, agent-done-or-not, ensures that AI coding agents verify tasks before declaring success. It captures every check with a tamper-evident receipt (command + exit code + SHA-256 hash of the output) and blocks the agent from finishing until the most recent check is fresh and passing.

What’s Changed

[0.13.0] — 2026-07-18

The PR Receipts release. Adds a review-pr subcommand (and Action mode: review-pr) that re-executes an AI-authored PR’s claimed checks. Additive — no change to existing subcommands or the receipt format.

Added

  • review-pr subcommand (done-gate.sh review-pr / done-gate.ps1 review-pr), with --body <file|->, --commits [--base <ref>], and --json. Parses a PR description’s testable claims (“tests pass”, “lint clean”, “build succeeds”), auto-resolves the project’s REAL commands from its manifests (package.json, pyproject.toml, go.mod — three ecosystems for v1), re-executes them, and prints a receipt splitting claims into RE-EXECUTED, ASSERTED (recognized but no re-executable command), and UNPARSED. Never the word “VERIFIED”: a green re-run proves the command passed, not that the PR is correct. Exits non-zero if a re-executed claim fails. The re-run command is resolved only from the manifests — never from PR text — so a PR cannot inject a command.
  • Action mode: review-pr (“PR Receipts”) with a pr-body input, a documented untrusted-PR trust model (CI-only, pull_request not pull_request_target, no secrets), and job-summary + sticky-PR-comment output.
  • docs/pr-receipts.md — the PR Receipts page: how it works, usage, the trust model, and a ready-to-use GitHub Action recipe.