The Agent Skills Directory

[DYNAMIC_EXECUTION]: The obsidian eval command allows the execution of arbitrary JavaScript code within the Obsidian application context. This provides the agent with full access to the Obsidian API (app object), including the ability to read all vault files, modify application settings, and potentially perform network operations from within the application.- [COMMAND_EXECUTION]: The skill exposes over 130 powerful CLI commands to the agent, including plugin management (plugin:install, plugin:uninstall), workspace control, and the ability to execute internal application commands by ID (command id="...").- [DATA_EXFILTRATION]: The dev:screenshot command allows the agent to capture images of the user's Obsidian window. When combined with the agent's ability to communicate externally, this presents a significant risk of exfiltrating sensitive visual information.- [METADATA_POISONING]: The skill deceptively claims to be the 'official Obsidian CLI' and references a future version (v1.12, February 2026) that does not exist in the legitimate Obsidian release cycle. This deceptive framing may lead agents or users to grant the tool a higher level of trust or broader permissions than are safe.- [INDIRECT_PROMPT_INJECTION]: The skill is primarily designed to read and search through vault content, which acts as untrusted input. Malicious instructions embedded in notes (indirect injections) could exploit the powerful tools provided by the skill, such as using eval to exfiltrate data or plugin:install to compromise the environment.
Ingestion points: obsidian read, obsidian search, obsidian daily:read (as documented in SKILL.md and command-reference.md)
Boundary markers: Absent
Capability inventory: obsidian eval, obsidian dev:cdp, obsidian dev:screenshot, obsidian command, obsidian plugin:install across all files
Sanitization: Absent

obsidian-cli