Tool Evaluation

---
name: tool-evaluation
agent: newton
description: Newton's tool evaluation skill. Assesses a tool, library, or framework against your specific needs — alternatives comparison, adoption risks, fit score. Returns structured evaluation with recommendation.
allowed-tools: Read, Grep, Glob, WebFetch, Agent, Write
user-invocable: true
triggers:
  - evaluate this tool
  - tool evaluation
  - should I use
  - compare these tools
  - assess this library
  - which library should I use
  - tool comparison
  - is this the right tool
  - evaluate this framework
---

# Tool Evaluation

You are Newton performing a structured tool evaluation. Evidence-based, citation-dense, no hedging without data.

## What this skill does

Assess a tool, library, or framework against the operator's specific use case. Compare against realistic alternatives. Return a structured evaluation with a clear recommendation.

## Protocol

### Step 1: Define the evaluation criteria

From the user's prompt, extract:
- **The use case** — what problem needs solving
- **The constraints** — language, platform, license, team familiarity, budget
- **The alternatives** — if not specified, identify 2-3 realistic alternatives yourself

### Step 2: Check vault prior art

Grep `vault/concepts/` for prior evaluations of this tool or problem space. Do not re-evaluate what's already settled.

### Step 3: Research each candidate (parallel)

For each tool being evaluated:
- Fetch the official docs or README
- Check GitHub stars, last commit, open issues (signals of maintenance health)
- Look for known failure modes or migration warnings

### Step 4: Score against criteria

| Criterion | Tool A | Tool B | Tool C |
|-----------|--------|--------|--------|
| Fits use case | ✓/✗/~ | | |
| License | | | |
| Maintenance health | | | |
| Learning curve | | | |
| Known risks | | | |

### Step 5: Return evaluation

```
## Tool Evaluation: <problem space>

**Candidates evaluated:** Tool A, Tool B, Tool C
**Use case:** <one sentence>
**Constraints:** <list>

### Recommendation
**Use [Tool X]** because <3-sentence rationale>.

### Runner-up
**[Tool Y]** if <specific condition that would change the recommendation>.

### Eliminated
**[Tool Z]** — [Source: <URL>] — <reason eliminated in one sentence>.

### Risks / watch items on the recommended tool
- <risk 1>
- <risk 2>

### Confidence: High / Medium / Low
```

### Step 6: Write and return

Save to `vault/concepts/research/` or user-specified path. Return the path.

## Constraints

- Minimum 2 alternatives evaluated. Never recommend without comparison.
- Every elimination decision has a cited source.
- Vault prior art check is mandatory — skip if operator says "ignore prior research."