Research to Rubric Colens Tutorial

Purpose

Learn how LenserFight represents a Colens as a local COLENS.MD automation object with ordered steps, lenser references, a Lens reference, and approval gates.

Concepts Covered

Colens, Lens reference, AI lenser, AI lenser team, approval gate, evaluation output, run report.

What You Will Build

You will simulate examples/colenses/research-to-rubric, a colens that turns a topic into a battle task and rubric.

Prerequisites

• Node 22.
• Dependencies installed.
• CLI built with pnpm nx run cli:build.

File Structure

examples/colenses/research-to-rubric/
  COLENS.MD
  README.md

Step-by-Step Walkthrough

1. Open COLENS.MD.
2. Inspect colens_type: evaluation.
3. Review the four steps: lens_execution, two agent_task steps, and an approval_gate.
4. Pass a topic through --inputs.
5. Inspect the generated runtime report paths printed by the CLI.

How to Run the Example

pnpm nx run cli:build
node dist/apps/cli/main.js colens run examples/colenses/research-to-rubric/COLENS.MD --inputs '{"topic":"local battle scoring"}'

Expected Output

Simulated colens Research to Rubric Colens
Status: ready
Steps: 4
JSON report: <user-runtime>/workspaces/<id>/runs/...
Markdown report: <user-runtime>/workspaces/<id>/reports/...

How the Example Works Internally

The CLI validates the automation object, reads frontmatter.steps, summarizes each step as <id> (<type>), and writes a local report. This is a simulation scaffold; it does not dispatch provider calls.

Common Errors and Troubleshooting

• Colens validation failed: check kind: colens, schema_version, id, and name.
• Invalid JSON supplied to --inputs: quote the JSON object correctly.
• No report appears: run from the repository root so .lenserfight/ is created where you expect.

Suggested Modifications

• Add a human_review_notes input.
• Change retry_policy.max_attempts.
• Add another agent_task for score aggregation.

Example Folder

examples/colenses/research-to-rubric