cowork-harness

Version updated for https://github.com/yaniv-golan/cowork-harness to version v1.1.0.

  • 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, cowork-harness, provides a scriptable, CI-friendly test harness to reproduce the observable runtime contract of Claude Cowork. It allows users to test their skills across many scenarios headless and in CI environments without relying on the locked Desktop app. The action supports different fidelity tiers, including replay mode for free demos, linting functionality with Python3, and live testing modes that require a running agent, token, and runtime environment. The summary highlights its main purpose, how it solves the need to test skills in various contexts, and the key capabilities it offers for developers.

What’s Changed

Minor: analyze-skill gains interactive-artifact write-back detection (static + an optional --runtime headless-DOM confirmer), lint gains a container-only-key tier check, and the doctor JSON envelope is frozen as a covered SPEC §12 surface. All additive.

Added

  • analyze-skill now detects interactive-artifact write-backs lost under Cowork. Alongside the existing /sessions path scan, it statically analyzes .html/.htm/.js/.mjs/.ts/.jsx/.tsx/.py sources under the target for a relative fetch/XHR/sendBeacon/<form method=post> write-back that silently fails under Cowork (the artifact is served from Cowork’s own origin, so a relative write-back resolves non-ok and a page that doesn’t check resp.ok shows a false “Saved”). Findings: artifact-write-back-lost (error — gates under --strict), artifact-write-back-suspect (advisory), and a separate top-level analysisFailures could-not-verify channel (a candidate that couldn’t be parsed/analyzed) that always exits 3, --strict-independent. A guard that isn’t statically provable-truthy is suspect, never silently clean; the blanket analyze-skill: ignore marker does not silence artifact rules. Each SkillFinding now carries a severity (error|advisory); --strict gates on any error finding.
  • analyze-skill --runtime — an optional headless-DOM confirmation that drives a materialized .html artifact in jsdom (stubbed network + synthetic user actions, run twice) to observe whether a relative write-back fires and is lost. Enrichment only (never changes the exit code); trusted-source scope. jsdom is an optional, dynamically-imported dependency — absent it reports “run npm i jsdom to enable”.
  • schema/doctor.json — the doctor --output-format json envelope is now a covered SPEC §12 surface (oneOf the completed-probe shape and the shared error envelope for every category). doctor’s normal JSON output is standardized through the shared envelope frame.
  • lint flags container-only assertion keys off-container — no_scratchpad_leak/present_files_called on fidelity: protocol|microvm|hostloop is an ERROR, on fidelity: cowork a WARN, clean on container.

What’s Changed

Full Changelog: https://github.com/yaniv-golan/cowork-harness/compare/v1...v1.1.0