The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing vendor-provided packages such as @inkeep/open-knowledge and @inkeep/open-knowledge-desktop via NPM. It also references downloading official DMG installers from the project's GitHub releases page. These resources are hosted on well-known, trusted platforms and originate from the skill's authoring organization.
[COMMAND_EXECUTION]: Instructions guide the agent to recommend or execute ok CLI commands, such as ok init, ok start, and ok open. These commands are specific to the OpenKnowledge platform and are required for its primary functionality of project management and file previewing.
[PERSISTENCE]: The ok init command is described as installing a project-local runtime skill into the agent's local skills directory (e.g., .claude/skills/open-knowledge/). This is a documented mechanism used to provide specialized behavior for the agent when working within an OpenKnowledge project.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to facilitate the reading and previewing of markdown files (.md, .mdx). This functionality introduces a potential surface for indirect prompt injection if an attacker-controlled file contains instructions meant to influence the agent's subsequent actions.
Ingestion points: Markdown and MDX files processed by the agent or previewed via the preview_url tool as described in SKILL.md.
Boundary markers: The skill does not explicitly define delimiters for external content, though it notes that runtime read/write rules are handled by a separate project-local skill.
Capability inventory: Command line execution of the ok utility and browser navigation through MCP tools.
Sanitization: No explicit sanitization or filtering logic for processed document content is mentioned in the discovery instructions.

open-knowledge-discovery