The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to install the defuddle package globally (npm install -g defuddle) if it is not present. This involves downloading and installing third-party code from the npm registry.
[COMMAND_EXECUTION]: The skill performs multiple shell operations using CLI tools like defuddle to parse web content and obsidian to search, create, and append data within the user's local vault.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it fetches untrusted content from external URLs and processes it without sanitization or boundary markers. This could allow an attacker-controlled website to inject instructions that influence the agent's note-taking or summarization behavior.
Ingestion points: Content extracted from external URLs via defuddle parse <url> in SKILL.md.
Boundary markers: None identified; the extracted content is directly interpolated into the note structure.
Capability inventory: Shell command execution (defuddle, obsidian), file system writes via obsidian create and the Write tool.
Sanitization: No sanitization or filtering of the extracted markdown content is performed before processing.

vault-capture