The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a strict priority hierarchy for its instructions, ensuring that repository-specific facts and AGENTS.md files override built-in rules. There are no attempts to bypass safety filters or disregard system instructions.
[DATA_EXFILTRATION]: No network operations (curl, wget, etc.) are present in the skill instructions. It focuses entirely on local repository analysis and content generation without external data transfer.
[REMOTE_CODE_EXECUTION]: The skill uses local markdown templates and rules to generate documentation. It does not download or execute remote scripts, and no unverified dependencies are introduced.
[COMMAND_EXECUTION]: While the skill uses system tools like 'rg', 'ls', and 'find' for repository grounding, these are used for read-only discovery purposes. File modifications are governed by a workflow that mandates explicit user confirmation before any write actions are performed.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from the target repository, creating a potential surface for indirect injection. However, the risk is mitigated by the lack of high-severity capabilities (such as network access or code execution) and the requirement for human review of all generated content.

doom-doc-assistant