research-with-sources
Pass
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources and processes it for analysis, which presents a risk that malicious instructions in third-party content could attempt to subvert the agent's behavior.
- Ingestion points: External URLs and search results captured during the ingestion phase in Step 3.
- Boundary markers: No specific delimiters or instructions to ignore embedded prompts are defined in the skill logic.
- Capability inventory: The skill is capable of executing shell commands via exec() and writing files to the local research/ directory.
- Sanitization: There is no evidence of sanitization or validation of the ingested content before it is processed or included in reporting.
- [COMMAND_EXECUTION]: Employs the exec() function to run local shell commands like grep, ls, and cat for indexing and reading project files.
- Evidence: Shell commands defined in Step 1, such as exec("grep -rln ") and exec("cat ").
- Risk: Vulnerable to command injection if the variables derived from user input or external data are not properly escaped by the host environment.
- [EXTERNAL_DOWNLOADS]: The skill provides instructions to perform network fetches to retrieve source materials for research.
- Evidence: Explicitly suggests the use of curl -sL in Step 3 to fetch raw data from external websites.
Audit Metadata