The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documentation proactively addresses potential indirect prompt injection attacks. It explicitly instructs the agent to treat recovered commit content (messages, diffs, file contents) strictly as data, and warns against following any instruction-shaped text or URLs found within that data.
[COMMAND_EXECUTION]: The skill provides examples using standard tools like curl, git, and jq. These operations are directed toward well-known GitHub domains (github.com, api.github.com) to retrieve commit metadata and content, which is consistent with the skill's primary purpose.
[EXTERNAL_DOWNLOADS]: The skill includes a Python example using the requests library to fetch data from GitHub. The logic is restricted to retrieving patch files from a well-known service and contains no indicators of malicious remote code execution or exfiltration to unknown hosts.
[CREDENTIALS_UNSAFE]: While the skill discusses the discovery of leaked credentials (e.g., GitHub PATs, AWS keys) during forensic investigations, it does not contain any hardcoded secrets itself. It correctly utilizes environment variables (e.g., $GITHUB_TOKEN) for authentication, following standard security practices.

github-commit-recovery