context-engine

Pass

Audited by Gen Agent Trust Hub on Nov 12, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill features tools designed to ingest and process external codebase data, which constitutes a potential surface for indirect prompt injection if the source files contain adversarial instructions.
  • Ingestion points: The scripts scripts/context_analyzer.py, scripts/context_pruner.py, and scripts/memory_indexer.py all read and process text from user-specified file paths.
  • Boundary markers: The tools do not currently implement specific delimiters or instruction-isolation markers in their outputs.
  • Capability inventory: The scripts/context_pruner.py utility includes file-write capabilities via the --output flag to save pruned content. No network or subprocess execution capabilities were identified.
  • Sanitization: The skill does not perform explicit sanitization or escaping of the content it reads.
  • [PROMPT_INJECTION]: A heuristic detector flagged a phrase in references/memory-and-context-editing.md as a potential injection. This is a false positive; the content is educational documentation that explicitly warns users about the dangers of poisoned memory files and provides best practices for securing persistent agent state.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 12, 2026, 07:06 PM