Test Function Quality Scorer

Internal scorer sub-agent that evaluates generated test function code using a 5-criteria quality rubric. Forked by test-plan.case-implement in Step 5.6.

Usage

This skill is not user-invocable. It is called by:

• test-plan.case-implement (Step 5.6) for quality assurance of generated tests

Inputs

From arguments

Parse $ARGUMENTS to extract:

1. --test-code-file: Path to file containing generated test code
2. --tc-file: Path to TC-*.md specification
3. --conventions-file: Path to repository conventions markdown
4. --framework: Test framework (pytest, Go testing, Jest, etc.)
5. --output-file: Path where assessment should be written (for revision feedback)
6. --calibration-file: Path to preloaded calibration examples from load_calibration.py (parent writes JSON .calibration_text to this file; this skill has no Bash). Read it as score anchors. Do not glob calibration/. Adding a pair is a new file under calibration/core/, calibration/ui/, or calibration/<team>/ — no SKILL.md edit.

Process

Step 1: Read Scoring Instructions

Read the detailed scoring rubric from ${CLAUDE_SKILL_DIR}/prompts/score-test-function.md.

Step 2: Execute Scoring

Apply the prompt with substitutions:

• {TEST_CODE_FILE} = --test-code-file argument
• {TC_FILE} = --tc-file argument
• {CONVENTIONS_FILE} = --conventions-file argument
• {FRAMEWORK} = --framework argument
• {OUTPUT_FILE} = --output-file argument
• {CALIBRATION_FILE} = --calibration-file argument (Read this file as score anchors)

The scoring rubric evaluates:

1. Coverage (0-2): All TC requirements implemented?
2. Assertions (0-2): Specific and meaningful?
3. Convention Adherence (0-2): Follows repo patterns?
4. Test Data (0-2): Uses realistic values from TC?
5. Code Quality (0-2): Clean, no excessive TODOs?

Step 3: Write Assessment to File

Write the structured markdown assessment to the --output-file path.

The output includes:

• Per-criterion scores with issues
• Total score (0-10)
• Verdict (Ready/Good/Revise/Rework)
• Coverage analysis (preconditions, steps, assertions implemented)
• Specific issues found
• Revision recommendations (if needed)

IMPORTANT: The orchestrating skill needs this file to:

1. Extract the verdict to decide if revision is needed
2. Extract specific issues to pass as feedback to the test generator
3. Keep an audit trail of quality assessments

What This Skill Does NOT Do

• Does NOT modify the generated test code (only assesses it)
• Does NOT trigger auto-revision (that's handled by the orchestrating skill)
• Does NOT make the revision decision (returns verdict, orchestrating skill decides)
• For scoring + auto-revision, the orchestrating skill (test-plan.case-implement) handles the revision loop

$ARGUMENTS