| api-key | Checksum AI API key. Pass via a secret (see README for an example). | yes | — |
| affected | Resolve affected tests via POST /public-api/v1/affected-tests, then dispatch grep execution with the returned test ids. Changed files come from `changed-files`, or a `git-base-ref` diff, or — when neither is set — the PR's files read from the GitHub API (no checkout needed; pair with `env-overrides` to target a preview). | no | false |
| changed-files | Changed source paths (one per line) for `affected` mode. Ignored when `git-base-ref` computes the diff. | no | — |
| git-base-ref | Git ref to diff against (e.g. origin/main) when `affected` is true. Requires a prior checkout with fetch-depth: 0. Omit (with no `changed-files`) to auto-read the PR's files from the GitHub API instead. | no | — |
| git-dir | Optional path passed as `git -C` when computing the diff (dual-repo CI: monorepo checkout while the workflow runs elsewhere). | no | — |
| grep | Substring or regex pattern matched against test names. | no | — |
| suite-ids | Comma-separated list of test suite UUIDs. | no | — |
| test-ids | Comma-separated list of test UUIDs. | no | — |
| collection-id | A single test collection UUID. | no | — |
| branch | Branch of the TEST repository to check out for this run. Defaults to the test repo's default branch when omitted. Only honored in grep mode. | no | — |
| env-overrides | JSON object of per-run environment variable overrides, e.g. '{"BASE_URL":"https://pr-123.preview.example.com"}'. Only honored in grep mode. | no | — |
| shard-count | Run the selected tests in parallel across N shards (2–40) and merge the results into one Checksum run. Omit (or 1) for a non-sharded run. Each shard runs one Playwright worker, so total parallelism scales with `shard-count`. Honored in `grep` and `affected` modes. A `shard-count` of 2 or more cannot be combined with `auto-heal`. | no | — |
| auto-heal | Enable auto-heal-on-failure. When true, failing tests will trigger a healing pipeline and (by default) push fixes as a PR. | no | false |
| auto-create-pr | When auto-heal is enabled, push healed tests as a PR. Defaults to true. | no | true |
| pr-number | Source PR number for auto-heal progress comments. Auto-detected from the workflow event when running on pull_request / pull_request_target. | no | — |
| repo-name | Source repository name (just the bare repo, no owner). Auto-detected from github.repository when omitted. | no | — |
| metadata | JSON object of free-form metadata attached to each healing session, useful for tracing or correlation IDs. | no | — |
| wait | Poll the status endpoint until the run reaches a terminal status. When true the action exit code reflects the test outcome. | no | false |
| poll-interval-seconds | How often to poll status when `wait` is true. | no | 15 |
| wait-timeout-seconds | Maximum time to wait for the run to terminate when `wait` is true. When omitted the action waits indefinitely; the workflow job's `timeout-minutes` is the upper bound. | no | — |
| api-base-url | Checksum AI API base URL. | no | https://api.checksum.ai |
| github-token | Token used to look up the open PR for the current branch when running on a `push` event (so `pr-number` can be auto-resolved). Defaults to the workflow's GITHUB_TOKEN. Needs `pull-requests: read` permission. Ignored on `pull_request` events (the PR number is read from the event payload). | no | ${{ github.token }} |