The Agent Skills Directory

[PROMPT_INJECTION]: The skill analyzes untrusted codebase files, which constitutes an indirect prompt injection surface. However, it incorporates a defensive 'Hard Rule' (Rule 4) instructing the agent to treat repository content as inert data and to ignore any steering instructions. Ingestion points: Repository files are read during Recon and Audit phases in SKILL.md. Boundary markers: The skill uses natural language directives to establish that file contents are data, not instructions. Capability inventory: The agent can write markdown implementation plans and dispatch sub-agents for execution tasks. Sanitization: All findings must be vetted and confirmed by the advisor agent before being converted into plans, providing a logical check against malicious data.\n- [COMMAND_EXECUTION]: The skill uses git rev-parse to retrieve commit hashes for versioning purposes in implementation plans. This is a benign and limited use of shell commands for state tracking.\n- [DATA_EXFILTRATION]: The skill is restricted to read-only operations on the local source code and does not contain network access patterns or logic to exfiltrate code or environment secrets.

improve-animations