docs-notes-retrieval

Pass

Audited by Gen Agent Trust Hub on Oct 6, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and package user-controlled markdown notes from a local vault into context for an LLM, creating a surface for indirect prompt injection.
  • Ingestion points: The scripts/scan_vault.py and scripts/build_context_pack.py scripts read arbitrary .md files from a user-specified directory path.
  • Boundary markers: The packaging script (scripts/build_context_pack.py) uses markdown headers (##), metadata blocks, and horizontal rules (---) to separate individual notes in the generated context pack. While these provide structural separation, they do not prevent an LLM from potentially following instructions embedded within the note content.
  • Capability inventory: The provided Python scripts perform file reads and write output to stdout or local files. The AI agent utilizing this skill likely possesses additional capabilities such as command execution, network access, and broader file system modifications, which could be targeted by an injection attack.
  • Sanitization: The skill does not perform sanitization, filtering, or validation of the text content within the notes to detect or neutralize embedded instructions before they are presented to the agent.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 6, 2026, 02:03 PM