The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to perform legitimate file management operations, including directory creation (mkdir), moving files to an archive (mv), and searching for references (grep, find). These actions are necessary for the skill's primary function and are performed with safety controls such as variable quoting.
[COMMAND_EXECUTION]: The orchestrator uses dynamic context injection (the !command syntax) in the SKILL.md file to gather live statistics, such as counts of markdown files and git history status. These commands are diagnostic, do not access sensitive data, and are used solely to provide context to the agent during initialization.
[COMMAND_EXECUTION]: The skill executes a local project script (bun meta-skills/scripts/manifest-sync.ts) to keep the artifact manifest synchronized with the disk state. This is a local workflow tool and does not involve downloading or executing remote code.
[PROMPT_INJECTION]: The skill reads manifest JSON files and markdown frontmatter from the workspace, creating an indirect prompt injection surface. This risk is effectively mitigated by the skill's rigid classification logic, the mandatory critic gate sub-routine that detects live references, and the requirement for explicit human confirmation for each category of file movement.
[SAFE]: No malicious patterns, such as obfuscation, credential theft, or unauthorized network communication, were detected. The skill prioritizes data integrity by never deleting files and protecting critical project folders (e.g., brand, architecture, research) from modification.

cleanup-artifacts