Skills
skills/jeremylongshore/claude-code-plugins-plus-skills/firebase-vertex-ai/Gen Agent Trust Hub

firebase-vertex-ai

Pass

Audited by Gen Agent Trust Hub on Apr 6, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in references/SKILL.full.md. User-supplied content from Firestore or HTTP requests is directly interpolated into prompts sent to the Gemini model.
  • Ingestion points: references/SKILL.full.md (functions analyzeContent and ragQuery use data.content and data.query).
  • Boundary markers: Limited to simple text labels (e.g., 'Content:', 'Question:'), which can be bypassed by adversarial input.
  • Capability inventory: The agent has Bash(cmd:*) and filesystem Write/Edit permissions, posing a risk if the agent's behavior is influenced by the processed data.
  • Sanitization: No input validation or instruction-filtering is implemented for the data processed by the LLM.
  • [COMMAND_EXECUTION]: The skill requires Bash(cmd:*) access to manage Firebase deployments, run emulators, and initialize projects via scripts/init-firebase.sh.
  • [EXTERNAL_DOWNLOADS]: The skill performs external downloads by installing the firebase-tools package globally via npm in scripts/init-firebase.sh and references/SKILL.full.md.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 6, 2026, 12:54 PM