The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it processes untrusted data from external sources, including Git repository metadata, GitHub API responses (specifically pull request comments and metadata), and Graphite CLI outputs. Instructions detail reading and writing PR observations to g2g/pull-requests.json and managing stack comments.
Ingestion points: Git repository state, gh API responses, gt (Graphite) CLI output, and g2g/pull-requests.json file.
Boundary markers: The skill instructions emphasize using structured data (JSON/porcelain) and specific schemaVersion markers for machine-readable output.
Capability inventory: The skill uses git, gh, gt, and brew via subprocess execution. It has file system read/write access and makes network calls via the GitHub CLI.
Sanitization: Instructions mention validating metadata and using structured templates for PR submission, though specific escaping logic for PR comments is not detailed.
[COMMAND_EXECUTION]: The skill requires the execution of multiple command-line tools to function. It uses git for branch operations, gh for GitHub interactions, gt for Graphite alignment, and brew for tool discovery and installation. These are necessary for the skill's primary purpose but represent a standard execution surface.

g2g