The Agent Skills Directory

[SAFE]: The skill is designed for autonomous codebase maintenance with significant emphasis on safety and verification. It includes explicit prohibitions against exfiltrating data, changing credentials, or performing destructive operations. While the skill interacts with untrusted repository content, creating a surface for indirect prompt injection, it includes mitigation strategies such as inspecting command wrappers and grounding all changes in concrete evidence. 1. Ingestion points: Repository instructions and local command definitions in the worktree (SKILL.md). 2. Boundary markers: No explicit delimiters are specified for untrusted repository data. 3. Capability inventory: Local file system write access, command execution (tests/lints), and subagent management. 4. Sanitization: The skill mandates inspecting wrappers before running commands and requires concrete evidence for all kept changes.

night-shift