The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions and templates for the agent to execute shell commands to run, lint, and capture output from generated source code in various languages (Python, Go, Rust, TypeScript, and C++). This execution is the core mechanism for verifying the 'concept bundles'.
[COMMAND_EXECUTION]: A Node.js script in SKILL.md (§15.2) performs a runtime smoke test by using the vm module's runInContext function to execute JavaScript code extracted from generated HTML files within a sandboxed context.
[EXTERNAL_DOWNLOADS]: The mermaid_validate.sh script and corresponding validation logic in SKILL.md (§15.1) utilize npx @mermaid-js/mermaid-cli to validate diagram syntax. This command downloads and executes the official Mermaid CLI from the npm registry.
[PROMPT_INJECTION]: The multi-agent workflow creates an indirect prompt injection surface as subagents (workers) ingest information from external web sources to generate code. The skill mitigates this through rigorous verification requirements.
Ingestion points: Worker subagents perform web searches (Step 2) to verify technical signatures and behavioral facts for the learning materials.
Boundary markers: The preamble mandates quoting authoritative sources and logging them in a _reference.txt file.
Capability inventory: The orchestrator agent environment contains standard compilers and interpreters used to execute and verify generated source files.
Sanitization: The skill relies on verification sweeps, lints, and 'pinned values' to ensure generated content remains consistent and correct.

concept-builder