The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub PR comments and review bodies to drive its decision-making logic.\n
Ingestion points: Untrusted content is fetched from GitHub API endpoints pulls/<pr>/comments and pulls/<pr>/reviews as described in the SKILL.md flow.\n
Boundary markers: While the skill utilizes a sub-agent context gate to triage inputs, it lacks explicit boundary markers or instruction-ignoring delimiters for the raw comment text during the grading phase.\n
Capability inventory: The skill has the capability to write local report files (.cheese/affinage/pr-<n>.md), post replies back to GitHub (post-reply), and dispatch tasks to the /cure skill, which is capable of performing code modifications.\n
Sanitization: There is no evidence of sanitization, escaping, or validation of the external comment content before it is processed by the LLM to generate recommendations.\n- [COMMAND_EXECUTION]: The skill relies on the execution of local bundled scripts and the GitHub CLI tool to perform PR operations.\n
Evidence: It invokes python3 skills/affinage/scripts/affinage.pyz with multiple subcommands (e.g., pr-status, age-route, post-reply) and uses gh api and gh pr view for interacting with the GitHub platform.

affinage