Skip to content

Add reproducible OpenSpec and project skill workflows - #12

Merged
kunal26das merged 1 commit into
mainfrom
feat/agent-workflows
Oct 14, 2026
Merged

kunal26das merged 1 commit into
mainfrom
feat/agent-workflows

Conversation

@kunal26das

@kunal26das kunal26das commented Oct 14, 2026 •

Copy link
Copy Markdown
Owner

Startup's maintenance guidance and release verification helpers were spread across documentation and temporary workspaces. This adds a reproducible project workflow for recording meaningful changes, selecting platform checks, preparing releases, and improving reusable skills.

  • Pin OpenSpec 1.14.1 with a locked, repository-local launcher and pure integration generation. Enable all twelve workflows for Codex and Claude, preserve user configuration, and detect generated-file drift.
  • Document five existing runtime capabilities and use OpenSpec to propose, verify, and archive this adoption change.
  • Add maintenance and skill-authoring skills. Retain the complete official Anthropic skill-creator at a pinned revision with its license and file hashes, alongside the OpenSpec license.
  • Add portable publication/wiki verification helpers, isolated evaluation fixtures, setup and contributor documentation, and Linux/macOS/Windows workflow CI.
  • Correct runner KDoc and examples to match existing behavior. Runtime implementation, public APIs, and the published artifact version remain unchanged.

Validation:

  • Locked npm installation, integration/provenance/link checks, six Node regression tests, and eight offline Python helper tests pass locally.
  • Existing JVM tests rerun: 106 Startup tests and 11 sample tests, zero failures/errors/skips.
  • Six current specifications pass strict validation; the completed adoption archive passes task validation.
  • Three paired maintenance-skill fixture cases were graded independently: candidate 15/15 expectations, baseline 14/15. This is a small smoke test, not an automatic-trigger or broad model-quality benchmark. Unsupported timing/token measurements were omitted. Details are in tools/ai/evaluations/startup-maintainer-2026-10-14.md.

Most added files are generated assistant integrations, mirrored skill resources, or unchanged upstream resources. Maintain the renderer and first-party .agents/skills/startup-* sources; regenerate their derived files with node scripts/setup-ai.mjs.

Known upstream dependency limitation: npm audit reports GHSA-vfj7-8cjw-p6xm in OpenSpec's glob dependency chain (one braces stack-exhaustion issue, propagated as four high entries). No patched braces release is available; the developer guide records this. The dependency is development tooling, outside published Kotlin artifacts. The suggested OpenSpec 0.17.2 downgrade would remove required workflow capabilities and was not applied.

Final verification at fd36803161b390b608320a1aa9ac08c00c8e4db0:

  • Agent workflow CI passed on Linux, macOS, and Windows, including all six Node tests, eight Python helper tests, provenance/drift checks, strict specs, and archive validation.
  • Existing library CI passed on Linux, macOS, and Windows.
  • A fresh staged checkout with CRLF conversion enabled passed all tooling checks; generated and vendored files retained LF. This was a local checkout-conversion check, in addition to actual Windows CI.
  • Anthropic's official quick_validate.py passed both first-party skills in an isolated local Python environment with PyYAML 6.0.3.

@kunal26das
kunal26das merged commit fd36803 into main Oct 14, 2026
6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant