The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends cloning several third-party community repositories and installing unvetted Obsidian plugins from individual developers (e.g., github.com/iansinnott/obsidian-claude-code-mcp, github.com/huytieu/COG-second-brain, and github.com/heyitsnoah/claudesidian). These sources are not from verified organizations and represent an unverifiable dependency risk.
[COMMAND_EXECUTION]: The documentation encourages the use of Git hooks (post-commit and pre-push) to automatically trigger claude CLI commands. While intended for automation, this setup creates a persistence mechanism where the agent autonomously performs maintenance tasks (like fixing links or updating logs) every time a user commits changes.
[DATA_EXFILTRATION]: The skill includes a 'Research Synthesis' and 'Article Import' workflow that fetches content from external URLs. This establishes a network-to-agent data flow that could be abused to exfiltrate sensitive vault information or ingest malicious instructions.
[PROMPT_INJECTION]: The integration relies on CLAUDE.md manifest files and vault-specific metadata (_meta/patterns.md) to define the agent's behavior and conventions. This allows the agent's operational logic to be influenced by the content of the files it is assigned to manage.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of untrusted data from the vault and the web.
Ingestion points: External article URLs (via references/workflows.md) and markdown content within the Obsidian vault.
Boundary markers: The provided prompt templates lack explicit delimiters or instructions to ignore embedded commands in the fetched content.
Capability inventory: The skill uses file-system read/write operations, network fetching, and shell command execution via the claude CLI.
Sanitization: There is no evidence of sanitization or content filtering before the external data is processed by the agent.

obsidian-claude-integration