The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves analyzing untrusted input such as code diffs and project documentation. This ingestion of external data is an inherent surface for indirect prompt injection, but the risk is managed through requirements for semantic review and verification of author-provided evidence.
Ingestion points: Processes code diffs, SKILL.md files, and Agent Notes (identified in SKILL.md).
Boundary markers: No explicit delimiter or sanitization instructions are provided within the skill text itself.
Capability inventory: The skill utilizes file system access and shell command execution (rg) to perform reviews.
Sanitization: None detected; the skill relies on the agent's internal safety filters and the human-in-the-loop review process described in the 'Reporting findings' section.
[COMMAND_EXECUTION]: A specific ripgrep (rg) command is provided for the agent to audit the workspace. This tool is used to identify non-standard or sensitive strings (e.g., local file paths like C:\ or /Users/, and platform names like OpenAI or Claude), serving as a security and portability check consistent with the skill's purpose.

workspace-standards-review