The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates JavaScript code at runtime within scripts/build_figjam_code.py by concatenating local templates found in the scripts/figjam/ directory with user-processed research data. This generated code is then executed using the use_figma tool to automate the creation of FigJam board elements.
[COMMAND_EXECUTION]: The script scripts/build_figjam_code.py uses the subprocess.run method to invoke the node binary with the --check flag. This is used to perform a syntax check on the generated JavaScript code before it is passed to the agent for execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data such as interview transcripts, chat logs, and desk research notes.
Ingestion points: External research sources provided by the user are processed by AI subagents (jtbd-extractor) as described in Stage 1 of the pipeline.
Boundary markers: The skill provides extensive documentation in the references/ directory, including strict schemas and extraction rules, to guide the AI and minimize the risk of malicious instructions within the source data influencing the output.
Capability inventory: The skill can execute local Python scripts, interact with the Figma MCP via use_figma, and manage files through the Google Drive MCP.
Sanitization: Extracted data is validated against a schema using scripts/validate_jobs.py, and the JavaScript generation script uses json.dumps() to ensure that data inserted into code templates is properly escaped.

jtbd-research-to-figjam