The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted data from git diffs and commit messages, which presents an indirect prompt injection surface. Mitigation is provided through explicit instructions in references/passive-data-safety.md to treat such data as passive and ignore embedded instructions. Ingestion points include git outputs and repository-based notes or specifications. Sanitization requirements include HTML escaping and secret redaction.
[COMMAND_EXECUTION]: The skill executes local git commands to resolve change ranges and gather commit information for analysis.
[DATA_EXFILTRATION]: The skill processes repository content that may contain sensitive data. To address potential exposure, the skill mandates the redaction of secrets, such as API keys and credentials, before they are included in the generated documentation.

explain-change