company-identity-reconciliation
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes a local bash helper script (
scripts/crawlora.sh) to perform API requests. The script is hardened by using the-qflag to ignore local curl configurations and includes strict validation of API paths and methods against an allow-list. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from web scraping and search results (ingestion points identified in
SKILL.mdandendpoints.md). The instructions mitigate this surface by requiring the agent to treat findings as attributed claims and to maintain strict separation between source data and canonical records. - [DATA_EXFILTRATION]: The helper script explicitly prevents local file disclosure by rejecting the
@character in query parameters, which blockscurlfrom reading local system files into the API request. - [CREDENTIALS_UNSAFE]: The skill handles the
CRAWLORA_API_KEYusing secure practices: it validates the key format and writes it to a temporary file with restricted permissions (chmod 600) that is automatically cleaned up on exit. This prevents the credential from appearing in process listings or being stored insecurely.
Audit Metadata