- Home
- Skills
- Code Quality & Review
- WHAT-IF — anchored premortem for your code
Works with the AI Agents you already use
WHAT-IF — anchored premortem for your code
Your code works on your machine. WHAT-IF shows you the week it stops working — N branches of your project's near future, each anchored to a line it actually re-read, each ending in one concrete fix.
Free
About this skill
WHAT-IF does what an experienced engineer does silently and a new developer can't yet: run the plan forward and notice what breaks.
Point it at a plan (before any code) or at code that already runs. It collects real anchors — a file and line it re-read, a plan item, a dependency version, a search that came back empty — and walks a fixed set of deviation operators (NONE, MORE, LESS, AS WELL AS, PART OF, REVERSE, OTHER THAN, EARLY/LATE) across them through up to 10 distinct lenses: error paths, omissions, volume, human behavior, external deps, money, time, concurrency, access, drift.
Each surviving branch is written as something that already happened:
### The same order was charged twice
Anchor checkout.ts:88 — <button onClick={submit}> with no disabled state
Lens / Deviation Human behavior / AS WELL AS
How Two requests left before the first response returned…
Cost Customer's money, a manual refund, a support conversation.
Early signal Pairs of near-identical orders — but nothing watches that table.
Window Cheap while the button is one component.
Action Disable the button until the response lands; send an idempotency key.
What makes it different:
- No anchor, no branch. Unanchored "the database might not scale" futures are deleted, not reworded.
- Honest count. Ask for 5, get "Requested 5, found 2" rather than padding.
- No fake probabilities. Branches are ordered by irreversibility: lost data → money → time → annoyance.
- Read-only. It never edits your code; the only file it writes is its own
.what-if.mdlog, so dismissed risks don't come back next run — unless the condition you dismissed them under has changed. - Built on measured results, not intuition: structured hazard questioning vs. brainstorming (Kobo-Greenhut 2019), method overlap of only 32–37% (Potts 2014), 92% of catastrophic failures from mishandled errors (Yuan, OSDI '14), omissions outnumbering commissions 2:1 (Bass, SEI 2006). Full references with DOIs in the GitHub README.
Not prediction: it widens coverage of concrete failure causes and nudges you toward reversible designs.
Usage: /what-if (3 branches) or /what-if 5. Source and plugin install: https://github.com/K0m04ek/what-if (MIT).
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 22 days ago
- Free with an Agensi account
Requires