result-diagnosis

Pass

Audited by Gen Agent Trust Hub on May 19, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it instructs the agent to ingest and analyze potentially untrusted data sources while maintaining access to high-capability tools like Bash and file system modification.
  • Ingestion points: The skill reads logs, configuration files, metric files, and checkpoints (references/evidence-audit.md), and fetches external content via web search and web fetch (SKILL.md).
  • Boundary markers: Absent. There are no instructions to use delimiters or specific warnings to ignore embedded instructions within processed data artifacts.
  • Capability inventory: The skill has access to Bash, Write, Edit, Read, and Glob tools (SKILL.md), allowing it to execute shell commands and modify the project environment based on its findings.
  • Sanitization: Absent. There is no mention of validating, filtering, or escaping content from logs or external documents before processing.
Audit Metadata
Risk Level
SAFE
Analyzed
May 19, 2026, 03:36 PM
Security Audit — agent-trust-hub — result-diagnosis