The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The troubleshooting documentation outlines an alternative installation method that fetches a shell script from an unverified third-party repository and executes it immediately via a pipe to bash (curl -fsSL "https://raw.githubusercontent.com/Dicklesworthstone/repo_updater/main/install.sh?ru_cb=$(date +%s)" | bash). This configuration introduces an untrusted remote code execution vector.
[EXTERNAL_DOWNLOADS]: The skill instructs users to download external scripts from an unverified third-party resource (Dicklesworthstone/repo_updater) instead of the declared vendor infrastructure or a recognized trusted service.
[INDIRECT_PROMPT_INJECTION]: The review system and automated agent-sweep workflows process data from external, potentially untrusted sources which exposes the execution environment to indirect prompt injection hazards.
Ingestion points: External GitHub issues, pull request content, and uncommitted workspace changes are processed dynamically in references/review-system.md and references/agent-sweep.md.
Boundary markers: There are no explicit text boundaries or instructions to isolate raw issue data or code contents from the agent prompt context.
Capability inventory: The underlying orchestrator possesses high-privilege capabilities including test execution (pytest, npm test), local script execution via hooks (pre_hook, post_hook), and direct repository modifications (git push).
Sanitization: No pre-processing sanitization, input filtering, or schema validation layers are documented for handling arbitrary issue descriptions or PR content.

repo-updater