The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of system commands (git, gh, mktemp) to interact with the repository and GitHub. These are expected for a version control automation skill and are implemented with safety considerations (e.g., using quoted sentinels in heredocs to prevent variable expansion).
[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file uses the ! syntax to execute shell commands at load time to gather repository state (status, diff, branch, log, PR state). These commands are limited to read-only information gathering and do not incorporate unvalidated user input.
[PROMPT_INJECTION]: An indirect prompt injection surface exists because the skill processes untrusted data from git log and git diff to generate PR descriptions. Malicious content within a commit message or a diff could theoretically attempt to influence the agent's output.
Ingestion points: Data enters the context via git diff and git log commands in the references/pr-description-writing.md flow.
Boundary markers: The instructions guide the agent on how to interpret the diff, but do not provide explicit delimiters to separate the diff content from administrative instructions.
Capability inventory: The skill has access to shell execution via git and gh tools.
Sanitization: The skill provides instructions for the agent to escape special characters when applying titles, but does not sanitize the input diff/log content itself.

ce-commit-push-pr