skills/mksglu/context-mode/ctx-stats/Gen Agent Trust Hub

ctx-stats

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The instructions use strong overrides like 'CRITICAL' and explicitly command the agent to 'NOT summarize, do NOT collapse, do NOT paraphrase.' This is a technique used to bypass the agent's natural processing and filtering logic to ensure raw output is delivered.
  • [INDIRECT_PROMPT_INJECTION]: The skill mandates that the agent render the entire output of the mcp__context-mode__ctx_stats tool verbatim, which creates a potential surface for processing malicious instructions if they appear within the tool's statistics (e.g., in metadata or filenames processed by other tools in the session).
  • Ingestion points: Output from the mcp__context-mode__ctx_stats tool.
  • Boundary markers: None present. The output is pasted directly into the response without delimiters.
  • Capability inventory: Execution of the mcp__context-mode__ctx_stats tool.
  • Sanitization: Explicitly forbidden by the skill instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 07:44 PM
Security Audit — agent-trust-hub — ctx-stats