Humans and AI use the same workstation¶
Scripts and profile commands are the supported interface. Documentation and Codex skills call those commands directly instead of maintaining a separate automation implementation.
The interface makes Windows diagnostic tools easier to find and operate. Humans can use the default output for direct troubleshooting. An AI can consume the structured form to handle slow or quirky evidence work, but it still reaches the same explicit privilege and state-change boundaries.
Contract¶
- Human-readable output is the interactive default.
-Jsonis the stable process boundary;-AsObjectis the PowerShell boundary where supported.- Inspection is read-only unless a command clearly says capture, enable, disable, ensure, reinitialize, copy, mount, or kill.
- Existing evidence is inspected before new evidence is collected.
- Privileged actions remain visible and explicit.
- Reports record the command recommendation and why it was selected.
Layers¶
| Layer | Purpose |
|---|---|
scripts/ and profile/ |
Atomic commands and implementation |
config/capabilities.psd1 |
Machine-readable capability and routing catalog |
tricky case schema |
Shared evidence boundary and reports |
docs/ |
Human workflows and reference |
.agents/skills/ |
Focused AI procedures built from the same commands |
| SkillOpt staging | Reviewed, validation-gated proposals for one skill at a time |
SkillOpt may improve descriptions and routing examples. It handles one skill at a time, stages its proposal outside Git, and validates the proposal against the command catalog. Adoption always requires a separate review and command.