Qodo published seven agent skills for the engineering work coding agents miss, and the list is notable for what it refuses to be about. Not one of the seven makes an agent write code faster. software-build-plan forces an evidence-backed plan naming scope, affected contracts, file responsibilities, test seams and failure behaviour before implementation starts. tdd-bdd turns a requirement into Given/When/Then and writes the failing test first. workflow-invariants models a multi-stage process as explicit states with allowed and blocked transitions. failure-path-testing proves a failed gate actually stops progression and that the operator gets a specific recovery action. Then three audits — dx-audit on changed commands, flags, output and errors from a second developer’s point of view, readme-audit on whether the README still matches the repository, and readme-creation for when it does not and you have authorised a rewrite. Their framing is the argument: “faster code generation is useful. A workflow that can explain, test, review, and defend the resulting change is what makes that speed sustainable.”
One design principle in there deserves to be a standard rather than a Qodo convention: “read-only and code-writing behaviour should always be visible before the skill runs.” Three of the seven are explicitly read-only audits, and knowing that before invocation is the difference between running a check and authorising a change. That is the same distinction an API expresses as a safe method or a dry-run mode, arriving in the skills layer because the skills layer now needs it. Their practical advice is also the right advice — “start with the place where your workflow currently loses confidence” rather than installing all seven.
The distribution behind this is not small. Qodo reports 903,500 installations of the VS Code extension and 648,500 of the JetBrains plugin, and the catalog carries seven Qodo API pages behind that reach: Qodo Merge (PR-Agent) is where a review skill would land, Qodo Gen and Qodo Command/CLI are the surfaces the Agentic Toolbox runs through, and the Qodo Hosted API is the programmatic door.
Qodo scores 31.6, thin on the Kin Score, with discoverability at 59.3 and access clarity at 47.4 against contract quality at 27.9 and contract governance at 0.0. Agent Readiness is 15.5, agent-aware, and the lit column has two entries in it — spec_presence and rate_limit_signal. Everything else is dark, including agent_skills, on the company that just published seven of them and named them precisely. dry_run_mode is dark too, which is the one that stings, because the read-only-versus-writing principle in that post is exactly what the dimension asks a provider to declare. A million and a half IDE installations is real distribution. None of it is reachable by anything that reads contracts rather than menus.