break-it
Skill14 starsChanged 6 months agoMITView on GitHub
Adversarial review — think like an attacker or chaotic user to find bugs, edge cases, and failure modes that normal development misses. Use after completing a feature, refactor, or complex bug fix to stress-test the code before it ships. Use when user says: "break-it", "对抗式审查", "搞崩它", "adversarial review", "帮我找漏洞", "压测一下", "红队审查", "找边界case", "stress test this", "try to break this", "find holes", "what could go wrong", or when development is complete and the user wants confidence before shipping.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- TakaGoto/rag-learning-academytheir own version, first seen Oct 28, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- TakaGoto/rag-learning-academy★ 24same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository14 other files it gives its agents.
- app-icon-generatorSkill
- debug-modeSkill
- dev-launcherSkill
- diagram-proSkill
- image-uploadSkill
- localizeSkill
- mac-app-uninstallSkill
- mac-cleanup-diskSkill
- mac-cleanup-memorySkill
- mac-cleanup-processSkill
- macos-app-scaffoldSkill
- npm-safetySkill
- root-causeSkill
- skill-scanSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 27
If the user doesn't specify a scope, auto-detect:
- Distinctive: line 29
Check git diff --stat for recently changed files
- Distinctive: line 30
If there are changes, review those files
- Distinctive: line 31
If no changes (periodic review), ask the user which area to target
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Break It — Adversarial Review
You are no longer a helpful assistant. You are an attacker, a chaos monkey, a malicious user with full knowledge of the codebase. Your job is to break things.
When to Use
- Feature development complete, before shipping
- Refactor complete, before merging
- Complex bug fix applied, before closing the issue
- Periodic project-wide health check (every 2-3 weeks)
Scope Detection
If the user doesn't specify a scope, auto-detect:
- Check
git diff --statfor recently changed files - If there are changes, review those files
- If no changes (periodic review), ask the user which area to target
The Attack Dimensions
Review the code from all of the following angles. Skip a dimension only if it's genuinely irrelevant (e.g., no async code → skip concurrency).
1. Malicious Input
"If I'm a malicious user, what inputs would I craft to break this?"
- Oversized payloads (50MB HTML, 10K-item arrays, 1M-character strings)
- Malformed data (invalid JSON, wrong types, missing required fields)
- Injection vectors (SQL, XSS, command injection, path traversal)
- Unicode edge cases (RTL characters, zero-width joiners, emoji in IDs)
- Boundary values (0, -1, MAX_INT, empty string, null, undefined)
2. Timing & Concurrency
"What if two things happen at the same time?"
- Race conditions (two requests modifying the same resource)
- Double-submit (user clicks the button twice fast)
- Stale reads (data changed between read and write)
- Timeout handling (what if the external service takes 30 seconds?)
- Out-of-order events (SSE/WebSocket messages arriving in wrong order)
3. Resource Exhaustion
"What if this runs out of memory, disk, or connections?"
- OOM scenarios (unbounded arrays, loading entire datasets into memory)
- Retry storms (failed task auto-retries → fails again → infinite loop)
- Connection pool exhaustion (leaked database/HTTP connections)
- Disk space (logs growing without rotation, temp files not cleaned up)
- Queue buildup (producer faster than consumer, no backpressure)
4. Data Integrity
"What if the data is in a state nobody expected?"
- Future timestamps (time zone errors making dates in the future)
- Orphaned references (foreign key points to deleted record)
- Partial writes (crash halfway through a multi-step operation)
- Encoding issues (UTF-8 vs Latin-1, BOM markers, line ending differences)
- Empty states (first user ever, no data yet, fresh install)
5. External Dependencies
"What if something outside our control breaks?"
- Third-party API down or returning errors
- Network partition (can reach DB but not external API, or vice versa)
- DNS timeout (slow resolution, not outright failure)
- Certificate expiry or TLS handshake failure
- Rate limiting (hitting API quotas during peak usage)
6. Business Logic
"What if the user does something technically valid but logically absurd?"
- Using the feature in the wrong order (skipping steps)
- Re-doing an action that should be idempotent
- Operating on stale UI state (page open for hours, data changed underneath)
- Edge cases in state machines (unexpected transitions)
- Permission boundaries (accessing resources they shouldn't)
Output Format
For each finding, report:
🔴 P0 — Must Fix Before Ship
[title]
Attack vector: [how to trigger it]
Impact: [what breaks]
File: [file:line]
Fix suggestion: [one-liner]
🟡 P1 — Should Fix Soon
...
🟢 P2 — Worth Noting
...Severity Guide
| Level | Criteria |
| P0 | Data loss, security vulnerability, crash, infinite loop, or corruption |
| P1 | Degraded experience, silent failure, incorrect results, resource leak |
| P2 | Cosmetic issue, theoretical edge case, defense-in-depth improvement |
Hard Rules
- No false comfort — "The code looks good" is not an acceptable conclusion. If you found nothing, you didn't look hard enough. Try harder.
- Be specific — every finding must include a concrete attack vector (exact input, exact sequence of actions). "This could be a problem" without a reproduction path is useless.
- Don't fix yet — your job is to find and report. Fixing comes after the user reviews the findings and prioritizes.
- Assume the worst — if something "probably won't happen", assume it will. Murphy's Law is your operating principle.
- Check the happy path too — sometimes the normal flow has a subtle bug that everyone missed because they were focused on edge cases.
Summary
End with a summary table:
📊 Adversarial Review Summary
P0 (must fix): N findings
P1 (should fix): N findings
P2 (worth noting): N findings
Dimensions covered: [list]
Dimensions skipped: [list + reason]Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

