The Agent Skills Directory

[COMMAND_EXECUTION]: The skill legitimately executes shell commands including git, python3, glab, and tox. These calls are scoped to repository management, validation, and merge request creation. The bundled Python script uses list-based subprocess calls to avoid shell injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill processes external YAML files which represents a potential attack surface, but it mitigates this through safe processing techniques. Ingestion points: YAML configuration files within the konflux-release-data repository. Boundary markers: Extensive use of dry-run validation and mandatory manual review of git diff before any publication. Capability inventory: Local file system writes, git commits, and merge request management on internal GitLab infrastructure. Sanitization: The bundled script uses line-by-line regex processing rather than a full YAML parser to avoid complex data injection attacks, and validates version strings against strict patterns.
[SAFE]: The skill incorporates multiple security safeguards, including verifying the target repository identity ('origin' URL) and checking for clean working trees before performing mutations. Atomic file operations with rollback logic prevent corrupted states during failure scenarios. The script also includes checks to prevent symlink attacks and escapes from the intended directory.

rhdh-konflux-rpa