The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to discover and execute arbitrary validation commands (such as test, lint, build, and type-check) found within the repository or recorded in the GitHub issue tracker.
Evidence: The SKILL.md file explicitly instructs the agent to discover project validation commands and run them during the start and implementation cycles.
[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by designating GitHub issues as the "Source of Truth" for the agent's planning and context.
Ingestion points: Data enters the agent context through gh issue view and gh pr view operations as defined in the SKILL.md instructions and references/templates.md.
Boundary markers: The provided templates for issues and trackers do not include explicit delimiters or instructions to ignore embedded malicious prompts within the processed text.
Capability inventory: The agent has the ability to execute shell commands (via the gh CLI and validation scripts) and manage repository state.
Sanitization: While the skill includes instructions to prevent shell injection when writing to GitHub by using files instead of inline strings, it lacks explicit sanitization or filtering for untrusted data read from issues that might contain adversarial instructions.

github-goal-delivery