The Agent Skills Directory

[METADATA_POISONING]: The skill-report.json file contains a pre-generated security report that claims the skill is safe, which is a deceptive practice intended to influence analysis tools. Additionally, the author listed in the metadata ('github') contradicts the author provided in the environment context ('aiskillstore').\n- [INDIRECT_PROMPT_INJECTION]: The skill uses self-referential metadata to attempt to dictate its own security verdict to analyzers.\n- [COMMAND_EXECUTION]: The documentation details commands for high-privilege operations, including repository deletion and administrative pull request merging.\n- [DATA_EXFILTRATION]: Instructions include commands for accessing sensitive authentication tokens (gh auth token) and listing repository secrets, which could be used for credential theft.\n- [REMOTE_CODE_EXECUTION]: The skill documents features for installing third-party extensions and executing commands in codespaces, representing surfaces for running untrusted code.\n- [EXTERNAL_DOWNLOADS]: Installation instructions reference fetching assets from trusted domains associated with GitHub.

gh-cli