obsidian-markdown

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes a local Python script (claude-obsidian.py) to manage vault operations. This is implemented using a secure transaction workflow where the agent must first generate and inspect a transaction bundle before applying it. The use of the --approved-plan-sha256 flag ensures that the agent can only execute changes that have been explicitly reviewed and hashed, preventing unauthorized or unintended modifications.
  • [INDIRECT_PROMPT_INJECTION]: As a tool for validating and drafting Obsidian notes, the skill naturally ingests untrusted user data.
  • Ingestion points: User-provided markdown notes and vault content processed for validation or synthesis.
  • Boundary markers: The skill uses YAML frontmatter boundaries and specific transaction bundles to isolate data.
  • Capability inventory: Execution of vault transactions via python3 "$CORE" transaction inspect/apply.
  • Sanitization: Includes instructions for YAML boundary consistency, deterministic wiki linting, and verification of internal targets and block references.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 05:51 PM
Security Audit — agent-trust-hub — obsidian-markdown