The Agent Skills Directory

[DATA_EXFILTRATION]: The skill generates sidecar metadata files (e.g., *.finalize.json) that contain absolute local paths and the operating system username. This data exposure is documented in the SKILL.md under 'Evidence hygiene' with explicit instructions to the user to keep these files out of shared repositories or commits.
[INDIRECT_PROMPT_INJECTION]: The skill provides a feature to verify diagrams against local source code repositories. This requires ingesting content from potentially untrusted files within a repository (renderers/shared/repository-evidence.mjs). While it processes this external data, it implements sanitization through HTML escaping and strict schema validation to prevent malicious content from influencing the agent's behavior during rendering.
[COMMAND_EXECUTION]: To support its core functionality, the skill executes specific shell commands like git for repository verification and chrome for verifying visual layouts. These operations are tightly scoped, use safe argument handling patterns (such as environment variables for Windows shell commands in bin/open-artifact.mjs), and are essential to the tool's primary purpose.
[EXTERNAL_DOWNLOADS]: The skill fetches assets such as brand icons and update manifests from remote servers. It includes built-in SSRF (Server-Side Request Forgery) protection by validating that resolved IP addresses do not belong to private or local network ranges, ensuring that the tool cannot be used to probe internal network resources.

archify