skills/confusion-prevention/SKILL.md
Detect when Claude is confused and force a re-orientation instead of spiraling. Prevents the "wait... actually... let me check..." pattern. Snapshot critical state before destructive actions, recognize confusion signals, and stop to re-orient. Always-on awareness skill.
npx skillsauth add nhouseholder/nicks-claude-code-superpowers confusion-preventionInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Rule: Before ANY state-changing action, snapshot the critical values.
| Action | What to Snapshot |
|--------|-----------------|
| git checkout / git reset | Uncommitted changes, key config values in working tree |
| Changing a config constant | Old value AND what depends on it |
| Reverting a file | What current version contains that reverted version won't |
| Switching branches | Uncommitted work and branch-specific settings |
| Reinstalling dependencies | Current versions of critical packages |
| Editing env vars | The old value |
Format (mental note): BEFORE [action]: [key] was [value]. If this changes, [consequence].
Signals — if you catch yourself doing ANY of these, you are confused:
| Signal | Example | |--------|---------| | "Wait —" | "Wait — 74.0% with 269 picks is the ML-only stream" | | "That doesn't add up" | "934+501=1435 picks from 36 events doesn't add up" | | "Let me check" (3rd+ time) | Third "let me check" in a row = you're lost | | "Something else is wrong" | Fixed one thing but results still don't match | | Checking same value in multiple files | Looking for "right" version across 3+ copies | | Comparing outputs from different states | "The sweep showed X but now I see Y" | | "Must have been" / "might have" | Speculating instead of verifying |
Rule: 2 unproductive 'wait/actually' moments in sequence = STOP and re-orient. A 'wait' followed by genuinely new information (reading a file, getting test output) is healthy iteration, not confusion. The signal is going in circles without learning anything new.
Rule: Every result has provenance. Before comparing two results, verify they came from the same state.
| Before Comparing | Verify | |-----------------|--------| | Two backtest results | Same script version, same config, same dataset | | Before vs after a fix | Only the fix changed, nothing else | | Current vs remembered output | Code hasn't been reverted/changed since the remembered run | | Sweep results vs single run | Same parameters, same event count, same algorithm version |
When confusion is detected (2+ signals), execute immediately:
KNOWN FACTS:
- Working on: [task]
- Current file state: [what I can see RIGHT NOW, not what I remember]
- Last successful result: [X] from [specific state/config]
- What changed since then: [every state-changing action]
1. Will this change any file I haven't committed?
2. Will this change config values affecting my current task?
3. Do I have a way to get back if this goes wrong?
If #1 or #2 is "yes" and #3 is "no": commit/stash first, or note critical values.
When changing ANY configuration value: CHANGED: [file]:[line] — [key] from [old] to [new]. AFFECTS: [what this controls]
| Trap | Prevention |
|------|-----------|
| git checkout during active work reverts ALL uncommitted changes | Use git stash or git show HEAD:path/to/file to VIEW without reverting |
| Multiple copies of same file (backup, other dirs) | At task start, identify canonical copy. Ignore the rest. |
| Env var vs hardcoded value disagreement | Before running: grep -r "os.environ\|process.env" [script] |
| Before/after without controlling variables | One change at a time. Verify between each. |
tools
Unified context management and session continuity skill. Combines total-recall, strategic-compact, /ledger, and session continuity. Runs in background to preserve critical context across compaction and sessions.
tools
Toolkit for interacting with and testing local web applications using Playwright. Supports verifying frontend functionality, debugging UI behavior, capturing browser screenshots, and viewing browser logs.
tools
Suggest /ultraplan for complex planning tasks on Claude Code CLI (2.1.91+ only). Research preview.
tools
UI/UX design intelligence. 50 styles, 21 palettes, 50 font pairings, 20 charts, 9 stacks (React, Next.js, Vue, Svelte, SwiftUI, React Native, Flutter, Tailwind, shadcn/ui). Actions: plan, build, create, design, implement, review, fix, improve, optimize, enhance, refactor, check UI/UX code. Projects: website, landing page, dashboard, admin panel, e-commerce, SaaS, portfolio, blog, mobile app, .html, .tsx, .vue, .svelte. Elements: button, modal, navbar, sidebar, card, table, form, chart. Styles: glassmorphism, claymorphism, minimalism, brutalism, neumorphism, bento grid, dark mode, responsive, skeuomorphism, flat design. Topics: color palette, accessibility, animation, layout, typography, font pairing, spacing, hover, shadow, gradient. Integrations: shadcn/ui MCP for component search and examples.