blog

v1.3 — The pipeline meets your PM tool

Editor’s note (June 6): skill and command names in this post predate the oc- rename — today’s /discover is /oc-discover, release-ops is oc-release-ops, and so on. Links point at the current pages. The history stays as written.

Until now, opchain and your project tracker lived in different universes. The pipeline would plan sprints, build them, and evaluate them — and then you’d alt-tab to Linear or Jira and transcribe what just happened like a court stenographer for your own robots. v1.3 closes that gap, and closes it in both directions.

PM-MCP runtime, in five skills

Five skills now speak to your PM tool at runtime over MCP — Linear, GitHub Issues, or Jira, resolved from one pm.yaml and a shared tool registry:

  • app-architect reads a ticket as discovery input (/discover --ticket ADEV-123) and writes sprint contracts back as comments — so the plan of record lives where your team already looks. (It’s the worked example; the same registry pattern serves the other four.)
  • On each sprint pass or fail, the build loop transitions the child ticket and posts the evaluator score. Your board reflects reality without a human retyping it.
  • Every write carries an idempotency marker, so re-running a phase never double-posts. Agents retry things; your ticket history shouldn’t read like a stutter.

The design principle: the PM tool is for state changes the team needs to see. Passing checks post nothing. Failures and contracts post once. Silence remains the default — we’d rather under-notify than train you to ignore us.

The platform menu

stack-forge breaks out of the JavaScript monoculture: the platform menu now spans Cloudflare, Django, Rails, Go, and Rust, each with real deploy patterns and test strategies rather than a paragraph of vibes. The stack decision was always the pipeline’s front door; now the front door has more than one hinge.

release-ops — the 18th skill, hired to ship itself

The new release-ops (/release) owns release cadence: it reads every skill’s checkpoint since the last release, proposes the next semver and theme, drafts the changelog from what actually shipped (not what we hoped would), bumps all skills in lockstep, and hands off to git-ops and deploy-ops.

Its first assignment was the release announcing its own existence. This very one. /release plan proposed “1.3.0”, drafted the entry you’ll find on the changelog, and walked the ship checklist — a new hire whose onboarding task was writing its own offer letter. It went fine, which is either reassuring or ominous, and we’ve decided to find it reassuring.

The through-line

Both halves of v1.3 are the same idea: the pipeline should push its state to the places humans already trust — the ticket board, the changelog — instead of hoarding it in a terminal scrollback. Checkpoints made the pipeline durable for the agent; v1.3 makes it legible to the team.

All 18 skills move to 1.3.0 in lockstep. Browse the skill library, or install and point /discover at the oldest ticket in your backlog. It’s not getting younger.

The opchain team

Builders of opchain

We build opchain — a skillchain and checkpoint protocol for shipping real software with Claude. We write about what we learn dogfooding it on our own pipeline.

← All posts