The Agent Skills Directory

[SAFE]: The skill implements a robust permission model. It explicitly requires the agent to seek human approval before executing commands involving network access, SSH, GitHub/registry authentication, or remote writes.
[SAFE]: Implementation of test fixes is strictly gated. The 'Implement mode' requires the agent to propose solutions and wait for explicit human agreement before making any changes to the codebase or updating the ledger.
[SAFE]: The skill logic for using sub-agents for code analysis is transparent. While it delegates review tasks, it maintains safety by requiring that any findings or actions resulting from that delegation are still subject to human-controlled approval gates.

test-gaps