Prompt StudioSAFE

CocoPlus is an Agentic Operating System for Snowflake Coco. It brings structured, multi-agent workflows to data engineering projects — covering everything from project initialization through spec, plan, build, test, review, and ship phases.

The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.

---
name: "prompt-compare"
description: "Compare two prompt versions by running both against the same test inputs and showing output differences. Usage: $prompt compare [prompt-a-path] [prompt-b-path]."
version: "1.0.0"
author: "CocoPlus"
tags:
  - cocoplus
  - prompt-studio
---

Your objective is to compare two prompt versions.

Before proceeding, verify that `.cocoplus/` exists.
If not: output "CocoPlus not initialized in this directory. Run `$pod init` to begin." Then stop.

Parse arguments: `$prompt compare [path-a] [path-b]`
If both arguments are missing: output "Usage: $prompt compare [path-a] [path-b]" Then stop.

If only one prompt path is provided:
1. Validate the prompt exists.
2. Offer to scaffold a draft comparison variant:
   "Only one prompt version was provided. Create an editable draft v2 for comparison? [yes/no]"
3. If yes, copy the prompt to the next version name in the same directory, add a clear frontmatter or top-of-file marker `draft_variant: true`, and add a note: "Draft variant scaffold — edit before using for final comparison."
4. Stop after creating the draft. Do not run `$prompt compare` until the developer has revised the draft.
5. If no, show usage and stop.

Read both prompt files. If either does not exist: output error with which file is missing. Then stop.

## Test Input Collection

Ask: "Provide 3-5 test inputs to compare the prompts against (one per line):"

## Run Comparison

For each test input, generate output using both prompts (invoke Cortex COMPLETE function via SnowflakeSqlExecute with each prompt as system context).

## Output Comparison Table

```
# Prompt Comparison: [prompt-a name] vs [prompt-b name]

| Test Input | Prompt A Output | Prompt B Output | Difference |
|------------|----------------|----------------|------------|
[one row per test input]

## Analysis
Prompt A: [strengths and weaknesses]
Prompt B: [strengths and weaknesses]
Recommendation: [which prompt to use and why]
```

## Anti-Rationalization

| Shortcut / Temptation | Why It Fails |
|-----------------------|--------------|
| Compare with only one input | A single case hides regressions and overfits the recommendation |
| Skip difference column details | Without explicit diffs, recommendation becomes subjective |
| Ignore missing prompt file validation | Comparison output becomes misleading and non-reproducible |
| Treat a scaffolded draft as a real strategy | The draft is only an editable starting point; the developer must revise it before comparison |

## Exit Criteria

- [ ] Both prompt files were validated as existing before comparison
- [ ] Comparison output includes rows for each provided test input with Prompt A output, Prompt B output, and a difference summary
- [ ] `## Analysis` section includes strengths and weaknesses for each prompt plus a recommendation
- [ ] Errors for missing arguments or missing files are handled with clear usage guidance
- [ ] One-prompt invocation can scaffold a clearly marked draft variant and stops before comparison

SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.

What the source does

Filesystem

none-observed

Network

none-observed

Shell

none-observed

Dependencies

pinned

Secrets in source

none-found

Findings (0)

No findings outside the package's declared scope.

Gates applied: no_behavioural_pass.