The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub issue bodies or local markdown files to drive automated code modifications and shell command execution.
Ingestion points: The agent fetches issue content using gh issue view or reads local files like docs/fitness-report.md (specified in SKILL.md).
Boundary markers: While the skill parses specific sections such as "Top 10 Action Items", it lacks explicit security delimiters or "ignore embedded instructions" warnings to prevent the agent from following malicious directives within the report content.
Capability inventory: The skill possesses extensive capabilities including file modification (Edit, Write tools) and shell command execution (Bash tool) for repository management and testing.
Sanitization: No sanitization or safety-filtering is performed on the external report data before it is interpolated into the agent's workspace logic.
[COMMAND_EXECUTION]: The skill executes various shell commands using git, gh (GitHub CLI), and uv (Python package manager). These tools are used to interact with remote repositories, manage issue statuses, and run test suites (pytest), with the specific parameters often derived from the triaged report data.

review-apply