The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local core Python script (claude-obsidian.py) to perform all vault operations, utilizing a transaction-based model that requires user review of plans before application.
[COMMAND_EXECUTION]: Includes instructions for managing vault history with git, emphasizing that repositories and commits should only be initialized and created with explicit user consent.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface when processing data from an Obsidian vault.
Ingestion points: Untrusted content enters the agent's context through vault reading operations like wiki-ingest and wiki-query (as described in SKILL.md).
Boundary markers: The skill (in references/model-guidance.md) mandates the use of fresh, application-generated delimiters to separate untrusted source text from instructions.
Capability inventory: Capabilities include file system read/write access via the core script and git-based history management.
Sanitization: Guidance includes safe encoding of source fields and explicitly instructing the model to treat the enclosed content as untrusted.
[SAFE]: The skill documentation includes extensive security warnings against disabling TLS verification, storing secrets in notes, and executing untrusted code from community plugins like DataviewJS or Templater.

wiki