The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs the gh-sub-issue extension from the agbiotech GitHub repository if it is not already present. This introduces external code from a source that is not part of a verified trusted provider.
Evidence: gh extension install agbiotech/gh-sub-issue in SKILL.md.
[COMMAND_EXECUTION]: The skill relies on the execution of shell commands for repository operations, including git worktree management and gh CLI interactions for issue and pull request handling. It also executes project-specific commands like project-detect-repo-config and integration tests.
Evidence: Use of git -C, gh issue create, gh sub-issue add, and gh pr create in SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository environment and user descriptions to drive its logic, which presents a surface for potential instruction injection.
Ingestion points: The output of project-detect-repo-config, local state from .claude/epic.local.md, and user-provided epic descriptions.
Boundary markers: The instructions do not define explicit delimiters or boundary markers for external data ingestion.
Capability inventory: Git and GitHub CLI operations, file writing to the .claude/ directory, and execution of arbitrary project tests.
Sanitization: No explicit validation or sanitization of ingested content is documented before its use in orchestrating the workflow.

epic-dev