Every repository has the PR. The body says “misc fixes,” the README describes a flag that was deleted in March, and the generated catalog is two skills behind the directory it catalogs. Nobody chose this; it’s just what happens when documentation is enforced by conscience. v1.8, Documentation & repo hygiene, replaces the conscience with a gate. Two new skills, and the catalog goes from 27 to 29.
The premise is the same one that runs through the whole pipeline: a rule that isn’t checked by something with the power to say no is a suggestion. oc-bug-check already applies that to code — no commit until the fast gate passes. v1.8 applies it to everything a PR carries besides code.
The two new skills
- oc-docs-forge (
/oc-docs) writes the documentation that must travel with a change. From the actual diff — not from what the author remembers changing — it produces the PR body’s## Documentationsection, a durable PR comment when the packet runs long, and edits to the README, product docs, changelog, and catalogs the change makes stale. Its hardest rule is about absence: when no user-facing docs changed, the packet must say so and argue why, with evidence. Silence is not a pass. - oc-repo-ops (
/oc-repo) is the repo-readiness gate. Before a PR opens, it verifies the docs packet exists and matches the current diff, generated files agree with their sources, catalog surfaces agree with each other, internal links resolve, no dirty files that look related to the PR are loitering untracked, and no checkpoint points at a file that no longer exists. It fails closed: warnings surface, failures block.
How they chain
The order is a dependency graph, and it runs on every PR:
- oc-docs-forge writes or verifies the documentation packet.
- oc-repo-ops verifies the packet and the repository around it.
- oc-bug-check runs the fast code gate.
- oc-git-ops opens the PR — only after all three pass.
The design rule underneath: the skill that writes the docs never decides whether they’re good enough. Author, verifier, and executor are three different skills with three different checkpoints. There’s a full engineering post on why that separation is load-bearing: The docs bot doesn’t grade its own homework.
Where the seams show (ours, not yours)
Two honest disclosures. First: like v1.7, this release ships contracts first — the two skills’ full command surfaces and cross-skill wiring land today, and parts of their reference guts harden over the coming days of dogfooding. Same seam two releases running; at this point it’s less a seam than a doctrine we haven’t written down yet.
Second: our own content calendar had v1.8 penciled in for July 29, conditional on there being a release at all. It’s July 10. The slip rule, it turns out, works in both directions — slots slip when the work isn’t worth shipping, and ship early when it is. The calendar has feelings about this; the calendar is not on the release train.
The through-line
v1.6 gave the pipeline eyes. v1.7 gave it the hard conversations. v1.8 makes every PR pay its paperwork before it’s allowed in the building — which is the least glamorous release we’ve cut, and the one whose absence you’d notice first, six months from now, reading a README that lies.
See the updated architecture diagram, browse the
29 skills, or install and run /oc-repo audit on
a repo you trust. Count the findings. We’ll wait.