The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection from the data returned by the external step.parts API.
Ingestion points: The agent reads part metadata (names, descriptions, and attributes) from the JSON responses of the /v1/parts and /v1/parts/{id} endpoints in scripts/download_step_part.py.
Boundary markers: Absent. The agent is instructed to directly evaluate and present search results to the user.
Capability inventory: The agent can invoke the bundled Python script which has capabilities to write files to the disk and make network requests.
Sanitization: Absent. No filtering or sanitization of the external API metadata is performed before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill downloads .step CAD files from the step.parts service.
Evidence: The script scripts/download_step_part.py fetches files from the stepUrl provided in the API response and saves them to a local directory (defaulting to /tmp/step-parts).
Mitigation: The script performs SHA-256 checksum verification to ensure the integrity of the downloaded data.
[DATA_EXFILTRATION]: The skill transmits user search queries and filtering parameters to an external service.
Evidence: User input is sent as URL parameters to https://api.step.parts. This is the intended behavior of the skill for searching the catalog.
[COMMAND_EXECUTION]: The skill utilizes a local Python script to perform its primary tasks.
Evidence: The agent is instructed in SKILL.md to run scripts/download_step_part.py for searching and downloading parts.

step-parts