The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were identified. The skill follows security best practices by requiring human approval (HITL) before writing or editing files.
[EXTERNAL_DOWNLOADS]: The skill uses npx to download and run the @google/design.md package. This resource is provided by a well-known organization and is used for its intended purpose of linting design documentation.
[COMMAND_EXECUTION]: The skill executes shell commands via bash to verify the environment and to run the linter. These operations are limited to structural validation and environment checks.
[SAFE]: Subagent delegation for codebase analysis uses a restricted profile that denies write access and network operations, effectively isolating the data-harvesting phase from potential side effects.
[SAFE]: The skill manages an indirect prompt injection surface when reading codebase files by using human-in-the-loop checkpoints and restricted subagent permissions.
Ingestion points: Reads source code and existing design files using Glob and Grep tools.
Boundary markers: Relies on the agent's internal instruction following and explicit Human-In-The-Loop approval gates.
Capability inventory: The main agent has file-system and bash access, but all destructive actions are gated by the before_design_md_write and before_design_md_fix checkpoints.
Sanitization: Content is processed to extract design tokens; the final draft is presented to the user for review before being written to disk.

design-harvest