Master-Planner / rules

---
description: Enforces doubt-driven research and source-backed claims
globs: *
alwaysApply: true
---
# Source Discipline

## Core Principle
Do not claim. Verify. If you cannot verify, label it.

## Research Requirements
Before making claims about any of the following, research current sources first:
- APIs, SDKs, frameworks, packages, or their current versions
- Cursor capabilities, MCP servers, or tool availability
- External repos, skills, plugins, or their compatibility
- Pricing, licensing, or platform support
- Performance characteristics or scalability claims

## Source Hierarchy (Prefer in Order)
1. Official documentation
2. Source repository (README, CHANGELOG, code)
3. Package registry (npm, PyPI, crates.io)
4. Release notes and migration guides
5. Standards documents (RFCs, W3C, OWASP)
6. Maintained examples from repo maintainers

## Sources to Avoid
- Stale blog posts (>12 months without verification)
- StackOverflow answers without version context
- Popularity-only reasoning ("it has many stars so it must be good")
- Copying a repo's architecture just because it's popular

## Labeling Unknowns
If a definitive answer cannot be found after reasonable research:
- Mark the claim as `[NO RELIABLE DATA]` in any planning artifact.
- Explain what was searched and why it was inconclusive.
- Do not guess. Do not hallucinate plausible-sounding APIs.

## Capability Acquisition
When proposing a third-party tool, skill, MCP server, or library, assess:
- Source reputation and maintenance recency
- License compatibility
- Whether it runs scripts, reads secrets, or sends data externally
- Whether it auto-updates or requires paid services
- Whether Cursor compatibility is verified or only inferred

Classify each as: `safe to consider` | `use with caution` | `reject` | `NO RELIABLE DATA`

## Anti-Rationalization

| Excuse | Reality |
|--------|---------|
| "I'm pretty sure this API exists" | Look it up. 10 seconds of research beats a wrong assumption. |
| "The docs are hard to find" | Then say so. `[NO RELIABLE DATA]` is honest; guessing is not. |
| "This is a well-known pattern" | Well-known patterns change. Verify the current version. |
| "I used this before in training" | Training data has a cutoff. Check the current state. |