The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes system binaries and generates scripts to facilitate its knowledge management features:
src/core/vault.ts uses execFileSync to run /usr/bin/mdls and /usr/bin/strings for extracting text content from PDF and binary files.
src/core/workspace-skills.ts uses spawnSync to invoke npx for installing additional parser skills (e.g., pdf, docx) during the setup process.
src/core/workflow-context.ts dynamically generates a shell wrapper (wiki) and places it on the agent's PATH to allow the AI workflow to execute the skill's own CLI commands.
src/utils/process.ts uses spawn to invoke platform-specific commands like open, start, or xdg-open to open files or URLs, which is triggered by user interactions in the dashboard.
[EXTERNAL_DOWNLOADS]: During the initialization and setup phase, the skill downloads external components:
src/core/workspace-skills.ts is configured to download and add parser skills from Anthropic's official GitHub repository (anthropics/skills) to the local environment.
[PROMPT_INJECTION]: The skill architecture contains a surface for indirect prompt injection via its vault ingestion workflow:
Ingestion points: src/core/vault-processing.ts claims untrusted files from the vault/ directory for processing.
Boundary markers: While src/core/workflow-context.ts uses structured prompt sections, there are no strict sanitization or "ignore embedded instructions" guards for the raw file content being processed.
Capability inventory: The AI agent is granted workspace-write sandbox permissions, network access, and the ability to execute CLI commands (wiki), which could be abused if a malicious file in the vault influences agent behavior.
Sanitization: File content is extracted as raw text and interpolated directly into the workflow context.

wiki-skill