The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill's documentation and installation scripts recommend a high-risk remote execution pattern: curl -fsSL https://raw.githubusercontent.com/akillness/obsidian-second-brain/main/scripts/quick-install.sh | bash. This allows for the execution of unverified remote code directly from a public repository.
[COMMAND_EXECUTION]: The skill relies on local bash scripts (scripts/install.sh, scripts/jeo-validate-ai-first.sh) for setup and vault validation. These scripts perform broad system modifications and configuration management.
[PERSISTENCE_MECHANISMS]: The scripts/install.sh script modifies the ~/.jeo/config.json configuration file to register a post-turn hook. This hook triggers the execution of scripts/jeo-validate-ai-first.sh every time the agent performs a file write or edit in subsequent sessions, establishing a persistent execution loop for all future agent interactions.
[INDIRECT_PROMPT_INJECTION]: The skill features a 'self-rewriting' vault logic that ingests untrusted external content and uses it to automatically rewrite existing notes, creating a large surface for indirect injection attacks.
Ingestion points: Commands like /obsidian-ingest, /x-read, and /research-deep ingest data from URLs, PDFs, audio files, and social media threads.
Boundary markers: The skill uses a ## For future Claude preamble and structured frontmatter, but lacks robust input sanitization or instruction isolation for ingested sources.
Capability inventory: The skill has access to Bash, Write, Edit, and WebFetch tools, which could be abused if an ingested source contains malicious instructions.
Sanitization: The skill relies on maintaining bi-temporal audit logs rather than active filtering or sanitization of ingested content.

obsidian-second-brain