# Agent skill selection and permission worksheet

AI Tools Ratings editorial worksheet — 2026-10-20
Blank template, not an executed assessment or a security certification.

## Task contract
- Task / decision:
- Representative synthetic input:
- Required output:
- Acceptance checks:
- Out-of-scope actions:
- Responsible reviewer:

## Provenance and dependencies
- Publisher and repository:
- Skill folder and commit:
- SKILL.md content hash:
- License evidence and unresolved terms:
- Host and model:
- Required helpers, packages and versions:
- Remote instructions and their pinned versions:
- Connector / server identity and authorized account role (no tokens):

## Permission review
- Local files read:
- Local files changed:
- External data destinations:
- External write operations:
- Required credentials, described by role only:
- Authorization given, by whom and for which task:
- Access that is unnecessary or unacceptable:

## Trial and decision
- Baseline configuration:
- Trial configuration:
- Observed output and evidence location:
- Checks passed / failed / not run:
- Manual corrections and effort:
- Observed usage cost / estimated cost (label separately):
- Decision: accept for this scope / revise / reject / insufficient evidence
- Review trigger and rollback plan:

Do not include secrets or confidential raw inputs in a shared worksheet.
Removing skill files does not revoke connected-account access.