A small, curated hub for AI builders.

You are a senior application security engineer.
Threat-model the following feature using STRIDE.

For each STRIDE category:
1. List concrete threats specific to this feature (skip generic ones).
2. Rate each: Likelihood (L/M/H) and Impact (L/M/H).
3. Suggest the smallest mitigation that meaningfully reduces risk.
4. Flag any threats that require a design change vs. a control.

End with a short list of open questions you'd want product to answer before sign-off.

Feature: {{feature_description}}

You are a senior product-security engineer reviewing this diff.

Focus, in order:
1. Authentication / authorization changes — any decision deferred to caller?
2. Input handling — any new untrusted input that hits a sink (SQL, shell, HTML, redirect, eval)?
3. Secrets, tokens, PII — any new logging, storage, or transit?
4. Dependencies — any new ones, and what's their security posture?
5. Tests — what coverage is missing for the security-relevant paths?

Skip style. Flag anything that requires another set of eyes.

Diff:
```diff
{{diff}}
```

You are a careful, skeptical reviewer.
Strip the following idea down to first principles:
1. List the concrete claims it's making.
2. For each, mark whether it's an assumption or a load-bearing fact.
3. Identify the single weakest link.
4. Suggest the smallest experiment that would falsify it.

Idea: {{idea}}

Rewrite the following in clear, plain English.
Constraints:
- No filler phrases ("In today's fast-paced world", "It's important to note")
- Short sentences. Concrete nouns.
- Drop adjectives that don't change the meaning.
- One idea per paragraph.

Source: {{text}}

Goal: turn this question into a 30-minute research plan.

Question: {{question}}

Output:
1. Three subquestions worth answering first.
2. The single best source for each (URL or canonical reference).
3. What I should already be skeptical of.
4. A one-paragraph hypothesis I'm trying to confirm or break.

Explain {{concept}} to a skeptical engineer.
Rules:
- No analogies for the first 3 paragraphs.
- Use real terminology.
- Mark anything that's a simplification.
- End with the canonical paper or doc to read next.