[AntoineMontane]

Adds a stackOutputs action output. In auto-approve-apply mode, the action now runs cdktn apply <stack> --auto-approve --outputs-file <file>, with the file under RUNNER_TEMP. After a successful apply it sets stackOutputs to the stack outputs as a JSON string.

Why: a common use is turning stack outputs into the job's deployment URL, or passing them to later steps:

environment:
  name: production
  url: ${{ fromJSON(steps.cdktn.outputs.stackOutputs || '{}').my-stack.url }}

Today that means parsing the output console text. hashicorp/terraform-cdk-action didn't support this either. Some forks added it by passing --outputs-file the same way.

Behaviour:

• Only auto-approve-apply writes the file and sets stackOutputs; the other modes are unchanged.
• If the apply fails, stackOutputs isn't set. If no file was written, it's {}.
• Sensitive outputs are left out, which matches the CLI default (--outputs-file-include-sensitive-outputs isn't passed).
• apply is an alias of deploy in cdktn-cli, which has the --outputs-file option.

Changes:

• engineCommand and buildNpxArgs take the optional outputs file;
• a new readStackOutputs helper;
• the output, in action.yml and a README row with a usage example;
• tests (4 new);
• rebuilt dist/.

npm run all passes: check, typecheck, 36 tests and the bundle.

🤖 Generated with Claude Code

[so0k]

Thanks for this, it's a useful addition. Your branch allows maintainer edits, so I've pushed a few commits on top rather than going back and forth on review:

• Renamed the output from outputs to stackOutputs (14fcc19). Every step output already lives under steps.<id>.outputs.<name>, so the old name gave workflows steps.cdktn.outputs.outputs, right next to the existing output (combined stdout/stderr) output, which made the two easy to mix up. stackOutputs matches the camelCase of exitCode. Since the name isn't released yet, changing it now costs users nothing; after a release it would be a breaking change. I've updated the PR title and description to match.
• Added run() tests (0aeaa0c) with @actions/core and @actions/exec mocked. They check that a successful apply publishes the outputs as compact JSON, and that a failed apply or a plan publishes nothing. The helper tests already covered the pieces; these prove the wiring in run().
• Comment and docs pass (eefbf88, 4981d53): tighter JSDoc, a note that cdktn-cli 0.24.0 and cdktf-cli 0.21.0 both support --outputs-file and leave sensitive outputs out by default, and README/action.yml now say the JSON is keyed by stack name and set only after a successful auto-approve-apply. dist/ is rebuilt with the lockfile's rolldown.
• Added a test-audit skill for agent-assisted test work (e28a81d). It's repo tooling only and doesn't affect the action.

[sakul-learning]

The .claude/skills/test-audit/ tooling added by @so0k currently mixes guidance for this GitHub Action repository with campaign instructions copied from the cdk-terrain monorepo.

In particular, CAMPAIGN.md directs agents to work in paths and harnesses that do not exist here, including packages/@cdktn/provider-generator, packages/cdktn, test/test-helper.ts, jest.preset.js, tools/build-test-matrix.mjs, and test/provider-tests/providers.json. It also tells agents to run pnpm package, while this repository is npm-managed (package-lock.json, no pnpm-lock.yaml). If Claude loads this skill during test work, those instructions would send it toward the wrong repository structure and commands.

Could we align the skill with this repository before merging it? Concretely:

• remove or rewrite CAMPAIGN.md around this action's actual src/, __tests__/, smoke fixture, and workflows;
• replace the remaining pnpm instructions with the repository's npm commands; and
• update TIERS.md where it still says the smoke test lacks coverage for the outputs output—the public output is now named stackOutputs.

The core stackOutputs implementation and its new run() tests look good; this feedback is limited to making the bundled test-audit skill safe and accurate for the repository in which it will run.