Pattern: Anti-rationalization (excuse/rebuttal tables)
Every process step an agent is tempted to skip is pre-empted by naming the rationalization (“I’ll add tests later”, “this change is too small to review”) and pairing it with a documented rebuttal. Coupled with a Red Flags list of warning signs that something has gone wrong. The technique hard-codes discipline into the workflow so the agent cannot quietly take the shortest path — the excuse it would reach for is already answered on the page.
This is the co-signature mechanism of two frameworks. addy-agent-skills puts the table in every skill; superpowers pushes it furthest, escalating the same instinct into Iron Laws (delete code written before its test; no fix without root-cause; no completion claim without evidence), per-skill Red Flags “STOP — you’re rationalizing” tables, and a bootstrap (sp-using-superpowers) that aims the technique at skill discovery itself. Both frameworks use it to make their skills resist the shortest-path default that motivates the pack.
Applied by (backlinks)
Addy Osmani — Agent Skills (every skill carries the table; representative):
- addy-tdd — rebuts “I’ll add tests later”.
- addy-code-review — rebuts “the change is too small to review”.
- addy-security — rebuts “no one will attack this”.
- addy-incremental-implementation — rebuts “it’s faster to do it all at once”.
Superpowers (Iron Laws + Red Flags + excuse→reality tables in nearly every skill):
- sp-using-superpowers — the Red-Flags table for skill discovery (“this is just a simple question”, “let me explore first”).
- sp-test-driven-development — the fullest excuse→reality table (“too simple to test”, “deleting X hours is wasteful” = sunk cost).
- sp-systematic-debugging — rebuts “quick fix now, investigate later” and “one more fix attempt”.
- sp-verification-before-completion — rebuts “should work now”, “I’m confident”, “agent said success”.
- sp-receiving-code-review — the social variant: forbids performative agreement (“You’re absolutely right!”).
See Also
- addy-agent-skills · superpowers — the two frameworks that standardize this across all skills.
- pattern-adversarial-review — a stronger, active form of the same distrust-your-own-output instinct.
- pattern-evidence-before-claims — the adjacent discipline: not talking yourself out of a process, but not declaring one finished without proof.