The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a bash helper scripts/crawlora.sh to interface with the Crawlora API. The script implements robust security measures, including restricted API key character sets to prevent injection into configuration files and strict path/method whitelisting that limits the agent's interaction to verified API routes.
[CREDENTIALS_UNSAFE]: The skill uses an environment variable CRAWLORA_API_KEY provided by the user. The helper script manages this safely by using a temporary curl configuration file with restricted permissions (chmod 600) and ensures the key is not passed as a command-line argument, preventing exposure in process listings or command history.
[DATA_EXFILTRATION]: The bash helper prevents local file disclosure and exfiltration by explicitly rejecting the @ character in query parameters and using standard input for request bodies, which prevents curl from interpreting user-supplied strings as local file paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from web searches and company reports.
Ingestion points: Bing search results, ImportYeti company reports, and scraped web content from official supplier websites as defined in SKILL.md and reference/endpoints.md.
Boundary markers: Instructions explicitly direct the agent to "separate claimed capabilities/certifications from independently verified facts," providing a logical framework for evaluating untrusted input.
Capability inventory: Shell execution is strictly limited to the scripts/crawlora.sh API helper; the skill lacks high-risk capabilities such as file system writes (except for a temporary credentials file) or dynamic code evaluation.
Sanitization: The crawlora.sh script sanitizes all inputs to prevent path traversal and prevents curl from accessing local files during API interactions.

supplier-sourcing-research