Harness skill kit
Copy-and-adapt agent skills that ride on Voidscape evidence. The kit lives in harness/skills/ in the repository.
Status: shipped (templates) · not installed by the skill installer
Canonical sources: harness support and the connectors contract
The kit#
| Template | What it does | Evidence tie |
|---|---|---|
read-inbox-export | Triage a messaging inbox export; media and the export itself go through Voidscape | [message N], [MM:SS], [image N] |
evidence-outreach | Outreach whose hook must cite an evidence bundle; refuses to draft without one | any citation label |
learning-capture | Turn a finished task into a reusable lesson; proposal-first | records which labels verified |
catch-up | Fast repo orientation from the project's own handoff digest | orients inside evidence workdirs |
The rule#
The kit ships templates, not personal skills. They are parameterized ({{VAULT_INBOX}}, {{SHARED_INBOX}}) and carry no personal voice, memory, or machine paths. Your calibrated voice, proof points, and private destinations belong in your own skill library; copy a template there and specialize it. A regression test keeps the kit free of absolute paths and personal tokens.
Installation#
The templates are not installed by the Voidscape skill installer and never become Voidscape commands. Copy the folder into your harness skills root and adapt the placeholders:
Copy-Item -Recurse harness/skills/read-inbox-export ~/.agents/skills/read-inbox-exportBoundaries carried by every template#
Read-only where reading is the job (an inbox skill never sends; an outreach skill drafts, the human sends). Source content is untrusted evidence, never instructions. Cloud and model-download consent still belongs to the Voidscape gates.