agentware: Skill for Claude Code

Environment Doctor

Portable Agent Skill (agentskills.io open standard). The YAML frontmatter above is the spec-compliant contract: name equals the folder name and description is the routing text every harness reads. The body is HARNESS-AGNOSTIC — no hardcoded invocation syntax (/skill, $skill), no harness-only frontmatter. Diagnosis runs read-only inside the workspace and needs no network; any fix that installs or mutates is PROPOSED, never auto-run.

When to invoke: when the environment itself — not the code you are about to change — is the blocker. A fresh clone won't build, a new machine is missing a toolchain, a runtime version mismatch breaks the install, an env var or service the app needs is absent, or a port the dev server wants is already taken. Run it FIRST when "it works on my machine" symptoms appear, before debugging the actual task. For a logic bug in code that builds and runs, use systematic-debugging instead; for dependency CVEs and supply-chain risk use dependency-supply-chain-audit.

Why this skill exists

agentware's whole ethos is verify-first: you cannot trust a build, test, or health command if the environment underneath it is wrong. A huge share of "the build is broken" time is actually environment drift — the wrong Node/Python version, a dependency installed at a different version than the lockfile pins, a missing DATABASE_URL, a port collision, or a service that never came up. These fail in confusing, misleading ways (a version mismatch surfaces as a cryptic compile error three layers deep), so the agent burns the failure-handling ladder on the wrong problem. This skill front-loads a deterministic environment diagnosis so the agent confirms the ground is solid BEFORE doing real work, and turns "it doesn't work here" into a specific, ranked list of fixes the operator can apply. It is the environment sibling of backend-verification and ui-verification: prove the prerequisites, don't assume them.