soia-pkm-maintain-vault-health

Pass

Audited by Gen Agent Trust Hub on Sep 1, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill implementation uses only standard Python libraries and lacks any network access or external dependencies.
  • [INDIRECT_PROMPT_INJECTION]: The skill reads untrusted content from a user's vault to audit links and tags.
  • Ingestion points: scripts/lint_vault.py and scripts/gen_vault_map.py read markdown and related files within the user-specified vault directory.
  • Boundary markers: Present; instructions specify that the scanning process is read-only and explicitly ignores system directories such as .git, .trash, and .obsidian.
  • Capability inventory: The skill can list directory contents, read file data, and write specific report files (e.g., OB知识库地图.md) to paths defined by the user or configuration.
  • Sanitization: Present; external content is parsed using regular expressions for specific structural metadata (wikilinks, tags) and is not executed or evaluated as code. The script gen_vault_map.py includes a symlink escape check to ensure output files remain within the vault boundary.
  • [COMMAND_EXECUTION]: The skill provides command-line interfaces for its Python scripts, but the code does not execute arbitrary system commands or inject user-supplied strings into shell environments.
  • [DATA_EXFILTRATION]: No network-related functions or data exfiltration patterns were detected. The skill is strictly limited to local filesystem operations.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 1, 2026, 03:52 AM
Security Audit — agent-trust-hub — soia-pkm-maintain-vault-health