The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a variety of shell commands and local scripts using data derived from commit messages and PR metadata.
Evidence: The skill uses git commit, git push, gh pr create, gh pr edit, and Sapling commands like sl metaedit and sl pr submit.
Arguments for these commands, such as --title "<title>", are directly populated from commit titles and PR views, which could lead to command injection if the agent's execution environment does not properly escape these variables.
The skill also executes local scripts including scripts/pr-info, scripts/pr-resolve-threads, and scripts/gh-upload.
[INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes data from external sources that can be controlled by third parties (e.g., via shared repositories or PR comments).
Ingestion points: The skill reads commit messages via git log -1 --format=%B (local but potentially from other contributors) and retrieves PR titles and bodies via gh pr view <number> --json title,body (remote content from GitHub).
Boundary markers: Absent. The skill does not define specific delimiters or instructions to the agent to disregard embedded instructions within the fetched commit messages or PR bodies.
Capability inventory: The agent has the ability to perform git push, create or edit GitHub PRs (gh pr create/edit), upload files (scripts/gh-upload), and resolve review threads (scripts/pr-resolve-threads).
Sanitization: Absent. There is no evidence of sanitization or filtering of the content retrieved from GitHub or local git logs before it is re-incorporated into new PR descriptions or command arguments.

submit