The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using uv run and the platform-specific orx skill command to vendor assets and run Python plotting scripts.
Evidence: uv run --no-project --with matplotlib --with numpy python figs/loss_curve.py and orx skill figures/assets/orx-chart.html > artifacts/orx-chart.html.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data (metrics and logs) which could potentially contain malicious instructions intended to influence the agent's behavior during figure generation or caption writing.
Ingestion points: The skill reads metrics from external CSV and JSON files identified via orx logs (e.g., figs/benchmarks.csv, figs/loss_curve.csv).
Boundary markers: Data is parsed using standard libraries (csv.DictReader, json.loads), but the agent uses this data to generate captions and summaries.
Capability inventory: The skill can write files to the workspace (artifacts/, figs/) and execute scripts via uv.
Sanitization: The provided Python templates perform basic sanitization by casting metrics to float and using json.dumps with character escaping for the HTML template.

orx-figures