The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill provides hardcoded default credentials (rustfsadmin) intended for local development environments and Docker containers running on localhost. These are documented in both the SKILL.md and references/demo-setup.md files for use with the RustFS storage backend.
[EXTERNAL_DOWNLOADS]: The skill clones the omnigraph-cookbooks repository from the vendor's official GitHub account to retrieve schema definitions, stored queries, and demonstration data.
[COMMAND_EXECUTION]: The skill instructs the agent to execute multiple system-level commands, including git clone, docker run, brew install, and various operations using the omnigraph CLI to manage cluster state and load datasets.
[INDIRECT_PROMPT_INJECTION]: The research workflow in references/research.md involves gathering information from external websites using WebFetch and WebSearch. This creates a vulnerability where malicious content on those sites could attempt to inject instructions into the agent's context during the signal extraction and pattern synthesis phases. (1) Ingestion points: External newsletters, blogs, and regulatory sites identified in the Phase 6 research workflow. (2) Boundary markers: The workflow does not specify the use of delimiters for external content, relying instead on a user review step of the intermediate seed.md file. (3) Capability inventory: The skill has the ability to write to the local graph (omnigraph load), modify configuration (omnigraph cluster apply), and launch a local web server (omnigraph-server). (4) Sanitization: While the skill mentions slug normalization and data deduplication, it lacks specific mechanisms for sanitizing natural language content against adversarial prompt injection.

omnigraph-intel-bootstrap