The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes Obsidian notes which may contain data from untrusted sources and has the capability to apply changes to the local vault.
Ingestion points: Processes user-requested Markdown notes and vault content.
Boundary markers: Relies on YAML property boundaries and explicit transaction review instructions.
Capability inventory: Modifies the filesystem via a local Python-based transaction tool (claude-obsidian.py).
Sanitization: Mandates a human-in-the-loop review process where users must inspect transaction bundles and provide a SHA-256 approval hash before changes are applied.
[COMMAND_EXECUTION]: The skill utilizes a specific local Python script to manage vault operations.
Evidence: Uses python3 "$CORE" transaction apply where $CORE is a script located within the skill's installation directory (/absolute/path/to/installed/claude-obsidian/scripts/claude-obsidian.py). This script execution is intended for the primary purpose of vault management.

obsidian-markdown