We shipped v1.6 yesterday. This is not a typo, and the blog’s date math is fine. Two releases in two days is either a process failure or a process working exactly as designed, and we’d like to argue — with a straight face — that it’s the second one.
Here’s the explanation. The whole catalog is versioned in lockstep: one changelog, one review, one bump moves all skills. That makes a release cheap, and when releases are cheap you stop batching them. Big releases are inventory, and inventory rots. v1.6 was the instrumentation release; the moment it landed, the next gap was obvious and already specced. So: v1.7, Seams & Signals — the seams between systems, and the signals that prove they work. The catalog goes from 24 to 27 skills.
The three new skills
- oc-signal-forge (
/oc-signal) turns a question into a trustworthy metric. Most dashboards are confident. Fewer are correct. The overlap is the product. signal-forge starts from the question a number is supposed to answer, designs the instrumentation, builds the harvester and transform — and then adversarially tries to refute the signal before it’s allowed anywhere near a chart. It closes a gap you might not have noticed in v1.6: oc-telemetry-ops meters the pipeline, oc-dash-forge renders beautifully, oc-monitoring-ops watches prod — and none of them owned the question “is this number true?” Now one of them does. - oc-modularize-ops (
/oc-modularize) decomposes a live monolith with provably zero functionality or data loss. The proof isn’t vibes: it captures golden fixtures from real traffic at every boundary and uses them as an equivalence oracle — the extracted module must replay identically before the cut counts. Its most senior-engineer feature, though, is refusal: if modularization isn’t actually warranted, it says so and stops. A skill that bills by the “no.” When the cut is real, it hands the bulk code-move to oc-migration-ops and per-module deployment to oc-fleet-ops. - oc-fleet-ops (
/oc-fleet) takes the territory oc-deploy-ops has always politely declined: self-managed infrastructure. Kubernetes, Nomad, Compose, on-prem VMs, Terraform and friends. It declares topology, provisions with the right IaC tool, rolls the fleet with an actual rollout strategy, and operates day-2. One rule is non-negotiable: a mandatory plan/dry-run gate before any IaC apply, becauseterraform applywithout a plan gate is a trust fall with your infrastructure.
deploy-ops and fleet-ops are peers, not a hierarchy: managed app → deploy-ops; self-managed fleet → fleet-ops. The platform matrix’s bare-metal row now routes accordingly.
How they chain
The point of a skillchain is that the pieces compose, so here’s the composed story. You have a monolith that’s earned a decomposition:
- oc-modularize-ops decides whether the cut is warranted (and is willing to say no), then captures golden fixtures from real traffic at each seam.
- oc-migration-ops executes the code move with rollback points.
- oc-fleet-ops lands the extracted modules on the fleet, plan-gated.
- oc-signal-forge builds the metrics that prove the migration did what it claimed — verified against the fixtures, not against optimism.
- oc-monitoring-ops enforces each signal’s freshness SLA from then on.
Where the seams show (ours, not yours)
Honesty section, as usual: a release this fast ships contracts first. The three skills land with their full command surfaces, architecture diagrams, and cross-skill wiring on day one; parts of their reference guts will harden over the coming days of dogfooding, with a checkpoint-protocol reconciliation sweep already queued behind them. We’d rather ship the contract and harden the internals in daylight than sit on a finished-looking release — but you deserve to know which one you’re getting on day one.
The through-line
v1.6 gave the pipeline eyes. v1.7 gives it the two hardest conversations in engineering: “this system needs to be cut apart” and “this number you love is wrong.” Both now have owners, evidence standards, and exits.
See the new rails on the architecture diagram, browse the
27 skills, or install and ask /oc-modularize whether
your monolith actually needs cutting. It might say no. That’s the feature.