The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external pull request comments and CI logs which could contain malicious instructions aimed at influencing the agent's behavior.
Ingestion points: Data enters the context through GitHub API calls for comments (pulls/<pr>/comments), review bodies (pulls/<pr>/reviews), and CI check summaries/logs as described in references/flow-details.md and SKILL.md.
Boundary markers: The instructions do not specify the use of strict delimiters or isolation techniques when passing raw, untrusted comment text to the reviewer sub-agent for grading.
Capability inventory: The skill possesses significant capabilities including posting replies to GitHub (post-reply), resolving PR threads via GraphQL, and triggering the /cure skill to perform automated code modifications based on the graded findings.
Sanitization: While the skill sanitizes PR references to extract integers, it does not mention sanitizing or escaping the content of comments or logs before they are processed or drafted into automated replies.
[COMMAND_EXECUTION]: The skill executes several local and external CLI tools to perform its tasks, including a bundled Python zipapp (affinage.pyz), the GitHub CLI (gh), Git, and specialized merge tools (mergiraf, kdiff3).

affinage