The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from GitHub Pull Request review comments, which could be crafted to include malicious instructions.
Ingestion points: scripts/fetch-comments.sh retrieves comment content via the GitHub API.
Boundary markers: The skill utilizes an adversarial verification process to challenge claims, though it does not explicitly sanitize comment text prior to processing.
Capability inventory: The skill can execute GitHub API writes (replies, thread resolution) and Git operations (commits, pushes, pull request creation).
Sanitization: A mandatory human-in-the-loop gate (AskUserQuestion) is enforced for every external operation, serving as the critical defense against prompt-based exploitation.
[COMMAND_EXECUTION]: The skill executes shell scripts and GitHub CLI commands to interact with the codebase and remote repository.
Evidence: The skill invokes gh api, gh repo view, and Git commands within its bash scripts.
Context: These operations are essential to the skill's primary function as a developer tool and are strictly gated by user confirmation before execution.

pull-spiderman