The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes content from external HTTPS URLs to generate Markdown for the agent to analyze. This creates an attack surface for indirect prompt injection, where a malicious website could include hidden instructions intended to override the agent's safety guidelines or influence its actions.
Ingestion points: Content fetched from the external HTTPS_URL variable in SKILL.md.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions for the extracted Markdown content.
Capability inventory: The skill can execute shell commands via python3 and an external defuddle binary, and it can write files to the .raw/captured/ directory.
Sanitization: The skill implements a 'Safety contract' that mandates URL validation (rejecting credentials, private hosts, and control characters) and prohibits shell interpolation by passing the URL as a single argv element.
[COMMAND_EXECUTION]: The skill triggers the execution of external binaries and Python scripts. It resolves the path for a core management script (claude-obsidian.py) and invokes an external extraction tool (defuddle). Although the skill includes guidelines for manual review of executable provenance and strict argument passing, the capability to execute arbitrary binaries based on remote input remains a sensitive security operation.

defuddle