The Agent Skills Directory

[PROMPT_INJECTION]: The skill has an indirect prompt injection surface through the ingestion of external GitHub issue data.
Ingestion points: Untrusted data is retrieved from external GitHub issues using the gh issue view command in Step 1.
Boundary markers: There are no explicit instructions to use delimiters or ignore embedded instructions within the fetched issue content when generating plans or code.
Capability inventory: The skill can perform file system writes (Steps 2 and 4) and execute shell commands (Step 6), which could be manipulated by malicious instructions in an issue.
Sanitization: Although Step 5 mandates a security review of the output code (OWASP check, secret detection), it does not sanitize the input issue text to prevent the LLM from following unauthorized instructions contained within the issue.
[COMMAND_EXECUTION]: The skill executes shell commands to interact with the environment and external services.
Evidence: Step 1 executes the gh (GitHub CLI) tool to fetch issue data.
Evidence: Step 6 executes arbitrary project-specific test commands which are determined at runtime by the project configuration.

implement-issue