# Assess Quality

> Evaluates execution outcomes against defined success criteria, scoring each criterion and producing a structured verdict with actionable feedback.

- Skill: `auto-skiller/assess-quality` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add auto-skiller/assess-quality`
- Raw SKILL.md: https://api.skillmd.com/api/skills/auto-skiller/assess-quality/raw
- Safety review: PASS (external: skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Tags: Evaluation, Feedback, Quality Assessment, Verdict
- Author: Auto-Skiller (https://skillmd.com/u/auto-skiller)
- Updated: 2026-08-22
- Page: https://skillmd.com/skills/auto-skiller/assess-quality

---


# Assess Quality

This skill allows the agent to rigorously evaluate the outcome of any executed step, comparing actual results against defined success criteria and producing a structured verdict with actionable feedback.

## Steps

1. **Establish success criteria.** Before evaluating, explicitly define what "good" looks like. If criteria are not provided, derive them from the goal context.
2. **Gather the output fresh.** Read or receive the actual execution output — do not evaluate from memory.
3. **Score against each criterion.** For each criterion, assign: ✅ Pass | ⚠️ Partial | ❌ Fail.
4. **List discrepancies.** For every Partial or Fail, describe exactly what is wrong and why it matters.
5. **Determine the overall verdict.** 
   - All Pass → **APPROVED** 
   - Any Partial (no Fail) → **CONDITIONAL** (list required fixes)
   - Any Fail → **REJECTED** (must be reworked before proceeding)
6. **Generate feedback.** For each non-Pass item, produce a specific, actionable fix instruction — not vague suggestions.
7. **Output** the full assessment.

## Output Format

```yaml
quality_assessment:
  subject: "<what was evaluated>"
  criteria_scores:
    - criterion: "<criterion name>"
      verdict: "pass | partial | fail"
      notes: "<observation>"
  overall_verdict: "APPROVED | CONDITIONAL | REJECTED"
  discrepancies:
    - item: "<what failed>"
      fix: "<specific action to resolve>"
  feedback_summary: "<one paragraph summary>"
```

## Notes
- Never emit APPROVED without checking every criterion.
- If success criteria are absent, generate them from context before scoring — never evaluate blindly.
- Pair with the `planning` toolbox to generate a remediation plan when verdict is REJECTED.