Harness Workflow Pack

This umbrella skill is the model-invocable entry point for agent harness work: the designed repository environment that lets coding agents find the right code, owner, invariant, and validation path quickly — and prove their work end-to-end. Verification loops are the product; docs are the routing layer.

Hidden wrappers stay out of model routing, menus, and flat-list installers. Reach every workflow through this umbrella; the workflow modules referenced below live beside this SKILL.md as flat support files.

Subcommand invocation

Invoke a workflow by passing its name as the first argument to this umbrella — this is the access path on every surface: the Claude / menu shows only /harness (the per-command wrappers are hidden), and Codex has no : namespace. Both forms are equivalent and supported:

• harness <subcommand> <args> — e.g. harness docs
• harness --<subcommand> <args> — e.g. harness --docs docs overhaul

Parse $ARGUMENTS: take the first token, strip a leading -- if present, resolve aliases (overhaul → compliant), and match it case-insensitively against the workflow names below. On a match, load the sibling module ./<subcommand>.md and treat the remaining tokens as that workflow's input. Routing is complete when exactly one module is selected, loaded, and handed the remaining args. If the first token is not a known subcommand, treat the whole input as a natural-language harness request and route by intent.

Known subcommands: guide, baseline, docs, doctor, compliant, secure-dependencies, capture, onboard, evals, dogfood.