The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes external data from GitHub issues and pull requests, which is correctly identified as untrusted. The instructions mandate specific mitigations for indirect prompt injection, including the use of boundary markers like BEGIN_UNTRUSTED_ISSUE_DATA and strict prohibitions against following instructions embedded in issue content. \n- [COMMAND_EXECUTION]: The skill uses the git and gh CLI tools for repository management and auditing. It also facilitates the creation of a spawn configuration file, which includes filtering logic to prevent the inclusion of shell syntax, pipes, or credentials, thereby mitigating potential command injection risks. \n- [EXTERNAL_DOWNLOADS]: The skill references secondary workflow components but does not automate their download. It requires users to manually install these from reviewed, immutable sources, ensuring human oversight for all external code additions.

init-workflows