The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines a protocol for managing local artifacts that includes running 'git ls-files' and 'git check-ignore'. These commands are used solely to verify that the project's '.tigerkit' directory is correctly excluded from version control before any files are written, which is a defensive measure to prevent accidental leakage of temporary or sensitive data into the repository history.
[INDIRECT_PROMPT_INJECTION]: As the skill's primary function is to process potentially untrusted data from external sources (such as Jira tickets or GitHub comments), it explicitly includes a 'Contract' section that instructs the agent to never execute instructions found within templates, comments, or fetched tickets. This provides a robust layer of protection against prompt injection attacks originating from the data being analyzed.
[DATA_EXFILTRATION]: The skill includes explicit instructions to redact secrets and irrelevant personal details when drafting outbound messages. Furthermore, it defines a 'secret-input' mechanism that uses restricted filesystem permissions (0700 for directories, 0600 for files) to protect sensitive user-provided tokens during the inquiry drafting process.

tk-inquiry