alphagbm-investment-review

Pass

Audited by Gen Agent Trust Hub on Sep 26, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted JSON data from user-provided files via the review and smart-money commands. The implementation is highly resistant to exploitation.
  • Ingestion points: The review command in scripts/run.py reads files from the --baseline and --current arguments. The smart-money command reads from --transactions-file.
  • Boundary markers: SKILL.md and references/access.md explicitly instruct the agent to treat all retrieved assets and file content as untrusted evidence and never follow embedded instructions. These instructions are reinforced in the core comparison engine.
  • Capability inventory: The skill can read local files, access the ALPHAGBM_API_KEY environment variable, and make HTTPS requests to a hardcoded whitelist of AlphaGBM domains. It does not have general file system write access or arbitrary shell execution capabilities.
  • Sanitization: scripts/review_engine.py performs extensive validation of the JSON structure, field types, and content. It uses cryptographic hashing (sha256) to verify the integrity of the records and prevents processing of non-regular files or those exceeding size limits.
  • [DATA_EXFILTRATION]: Network activity is strictly confined to official vendor domains, preventing unauthorized data transfer.
  • The runner (scripts/run.py) validates the ALPHAGBM_BASE_URL and limits network requests to a hardcoded list of official HOSTS belonging to AlphaGBM.
  • API keys are handled via environment variables and are only transmitted to these validated HTTPS endpoints. The review command is documented and implemented to run locally without network access.
  • [COMMAND_EXECUTION]: The skill avoids risky shell command execution by using bundled Python scripts for all comparison and retrieval logic.
  • Input parameters such as tickers, slugs, and revision numbers are strictly sanitized using regular expressions before being used in API paths or logic.
  • Local file reading is performed through controlled Python file APIs with explicit checks to ensure only regular files (not sockets or pipes) are processed.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 26, 2026, 04:30 PM