The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources such as Jira tickets, conversation histories, and user-defined templates. While it includes explicit instructions to ignore any embedded commands within this data, the ingestion of external content constitutes a vulnerability surface for indirect prompt injection.\n- [COMMAND_EXECUTION]: The skill uses local shell commands (git ls-files, git check-ignore) to manage its artifact directory .tigerkit/. These commands are used to verify repository state and ensure files are correctly ignored by version control before writing.\n- [DATA_EXFILTRATION]: Instructions in artifact-paths.md define a secure handling procedure for temporary files, including the use of restricted filesystem permissions (0600/0700) for "secret-input" directories. Additionally, the skill is instructed to redact personal details and secrets from generated drafts.

tk-to-questionnaire