The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to execute env | grep -i _TOKEN=. This pattern retrieves all environment variables containing authentication tokens and exposes them to the model's context. While intended for troubleshooting credential conflicts in the GitHub CLI, this broad grep command reveals sensitive secrets that are unrelated to generating pull request descriptions.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize data from sources that may contain untrusted content.
Ingestion points: Git commit messages (git log), diff statistics (git diff), and external Jira issue summaries fetched via the Atlassian MCP.
Boundary markers: The instructions do not define boundary markers (such as XML tags or delimiters) to separate external data from the skill's own instructions, nor do they include explicit 'ignore embedded instructions' directives for the agent.
Capability inventory: The skill possesses the ability to execute shell commands (git, gh, jq) and perform write operations to the repository via the GitHub API (creating and updating pull requests).
Sanitization: The skill effectively mitigates direct command injection risks by using jq with --rawfile and --arg flags to safely handle variables before they are passed to the gh api command.
[COMMAND_EXECUTION]: The skill performs extensive interaction with the local system via shell commands to gather git metadata. It follows security best practices by implementing a mandatory human-in-the-loop confirmation step before any write operations are executed and by ensuring variables are safely wrapped when passed to CLI tools.

generate-pr-description