The Agent Skills Directory

[DYNAMIC_EXECUTION]: In SKILL.md (Step 5), the skill uses the eval command to execute a git diff operation. This command interpolates variables such as $BASE_BRANCH and $GIT_DIFF_EXCLUDES. While these are intended to be branch names and pathspecs, using eval on strings that can be influenced by user or orchestrator input is a high-risk pattern that can lead to shell command injection.- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the summarization of code changes by an AI subagent, which creates an ingestion point for untrusted data.
Ingestion points: SKILL.md captures a git diff into a patch file.
Boundary markers: The instructions for the subagent in references/pr-body-summariser-prompt.md do not utilize robust delimiters or explicit 'ignore embedded instructions' warnings for the content of the diff file.
Capability inventory: The subagent's summary is used to generate pull request descriptions via gh pr create or the Bitbucket API.
Sanitization: The skill does not perform sanitization of the diff content before processing, allowing potential malicious comments in the code to influence the subagent's output.

create-pr