The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from multiple untrusted sources to automate issue creation and release documentation.
Ingestion points: Step 4 accepts sweep reports, scanner outputs, and user-pasted findings (SKILL.md). Step 10 aggregates titles from closed issues via the gh issue list command to generate changelog entries (release-pipeline.md).
Boundary markers: The skill utilizes predefined markdown templates (references/issue-body-template.md) and idempotency markers (e.g., ) to structure content, but lacks technical isolation for external strings.
Capability inventory: The skill possesses extensive capabilities including file modification, authenticated command execution (gh, git), and subagent orchestration (Task).
Sanitization: No automated sanitization or escaping of ingested text is implemented; the process relies on a mandatory 'preview → confirm' cycle (Step 7) for human verification of all generated content.
[COMMAND_EXECUTION]: The skill performs broad administrative operations on local and remote repositories using Git and the GitHub CLI.
Evidence: Operations include authenticated API calls (gh api), resource creation (gh issue create), and repository modifications like git push --force-with-lease (SKILL.md, pr-branch-conventions.md).
These operations are scoped within the skill's intended purpose of release management and follow strict human-in-the-loop confirmation requirements.
[DYNAMIC_EXECUTION]: The skill orchestrates complex, parallel workflows by dispatching tasks to multiple subagents.
Evidence: Step 9a describes a 'parallel dispatch' model where 10+ independent issue workers can be launched simultaneously using the Task tool to handle code changes (SKILL.md).

ship-cycle