The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub issue bodies and comments via gh issue view and gh api calls. This content is used to drive implementation decisions and code generation.\n
Ingestion points: SKILL.md (Step 1: Read the ticket and claim it) and references/auto-pick.md (Auto-pick a ticket).\n
Boundary markers: references/auto-pick.md includes logic to strip fenced code blocks before regex matching, but SKILL.md lacks explicit structural delimiters or instructions to ignore embedded commands when processing the full ticket body during implementation.\n
Capability inventory: The skill can execute arbitrary shell commands discovered in the repository (SKILL.md Step 4), perform file system modifications (Step 5), and mutate external state via the gh CLI (creating branches, PRs, and comments).\n
Sanitization: No specific sanitization or filtering of the ticket body content is performed before it is presented to the model as requirements for code generation or command execution.\n- [COMMAND_EXECUTION]: The skill identifies and executes shell commands derived from the repository's configuration files (e.g., Makefile, package.json, pyproject.toml) as specified in SKILL.md Step 4 and Step 6. This includes linting, testing, and application startup commands.\n- [DYNAMIC_EXECUTION]: The skill performs dynamic discovery of execution paths at runtime. It includes a 'Runtime smoke check' (Step 7) that starts applications in the background, waits for network ready signals, and exercises endpoints, involving process management and interaction with the running system.

implement-ticket