review-deck

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from user-provided PDF and PPTX pitch deck files to generate investment feedback.
  • Ingestion points: SKILL.md Step 2 uses the Read tool to load content from the file path provided in $ARGUMENTS.
  • Boundary markers: The skill does not implement explicit delimiters or instructions to ignore embedded prompts within the deck content.
  • Capability inventory: The skill is configured with Bash, Read, Grep, and Glob capabilities.
  • Sanitization: No sanitization or validation of the ingested text is performed before it is used for evaluation.
  • [COMMAND_EXECUTION]: The skill uses the Bash tool for internal directory discovery.
  • Evidence: SKILL.md Step 4 directs the agent to list directories under ${CLAUDE_SKILL_DIR}/perspectives/firms/ to identify available VC firm profiles.
  • Context: This execution targets a fixed internal path and does not involve user-controlled input, reducing injection risk.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 06:03 AM
Security Audit — agent-trust-hub — review-deck