deep-company-series
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest large amounts of untrusted external data, which creates a surface for indirect prompt injection attacks.
- Ingestion points: Processes 5 years of annual reports, quarterly reports, and third-party research reports via 'Read' and 'WebSearch' capabilities.
- Boundary markers: While the skill provides rigorous content templates and a fact-checking checklist, it lacks technical delimiters or explicit instructions for the agent to ignore embedded commands within the ingested research material.
- Capability inventory: The skill utilizes file writing, shell command execution ('Bash'), and Git operations ('git push').
- Sanitization: The skill relies on human review and cross-referencing for data accuracy but does not specify technical sanitization or escaping for data interpolated into agent prompts.
- [COMMAND_EXECUTION]: The skill instructs the agent to execute several shell commands as part of its standard workflow.
- Evidence: Uses 'date' to establish a temporal baseline, 'grep' to scan for PII (a security-positive action), and 'git push' to publish reports to a remote repository.
- Context: It also calls a project-specific tool, 'python3 tools/financial_rigor.py'. These commands are consistent with the skill's stated purpose of publishing research from a developer-like environment.
Audit Metadata