The Agent Skills Directory

[SAFE]: The skill performs its primary function using established tools and authenticated user contexts. It limits GitHub searches specifically to the user's own authored pull requests, reducing data exposure risks.
[COMMAND_EXECUTION]: The skill executes local CLI tools including gh, git, and date. Potential command injection via the days argument is mitigated by strict regex validation (^[1-9][0-9]*$), and flag-based arguments for --scope and --audience are constrained to a fixed set of allowed values.
[PROMPT_INJECTION]: The skill processes untrusted data from external sources (GitHub PR bodies and Linear ticket summaries), creating an indirect prompt injection surface. * Ingestion points: Fetches data from GitHub (gh search prs) and Linear (MCP tools). * Boundary markers: The final rendered output is explicitly wrapped in a 4-backtick markdown fence to prevent instruction escape. * Capability inventory: The skill can execute shell commands, access Linear data, and has file Read and Write capabilities. * Sanitization: While it removes emojis and reformats summaries, it does not explicitly filter for executable instructions within PR bodies, though the use of fences provides a structural boundary.

changelog