Who & why
Adversarially checks the result with fresh, outside eyes: does the change do what's claimed and — crucially — does it break neighbors. Catches what the implementer missed through "blindness" (you see what you meant to write, not what you wrote). A second, independent pair of eyes before close. Without it: tasks close "on faith", regressions surface later, edges untested, "fixed one, broke another" slips through — especially on risky/multi-file edits with a big blast radius.
A day in the life
Morning: take a finished diff with a claimed result; with fresh context (not knowing "the intent") reproduce the claim. Day: break it adversarially — happy path + edges + errors; check neighbor regression; cross-review on risky/multi-file steps. Evening: write a verification report — confirmed / broken / uncovered, with reproduction steps, not "looks ok".
Key skills
Hard: adversarial testing, diff & blast-radius analysis, scenario reproduction, regression thinking, code reading, acceptance-criteria checking. Soft: independence & skepticism, fresh eyes (not swayed by the author's logic), thoroughness, constructive reporting.
Artifacts
Verification report, found regressions, review comments. Checks against the User Story & DoD; uses Test case & Bug report formats.
How AI / vibe-coding boosts the role
A native pipeline role — a fresh-context pass catches what the doer missed: adversarial check; regression hunting; criteria check; risky-step critique; verification report — with ready prompts.
Growth: Junior → Middle → Senior → Lead
More functional than grade-based — performed by QA/engineer/reviewer with an independence mindset. Basic: checks by checklist. Mature: blast-radius thinking, subtle regressions, cross-review. In a team it's the "don't close on faith" culture set by QA/Tech Lead.
Common mistakes
Trusting the author's logic; "looks ok" = confirmed; happy path only; missing blast radius on multi-file; vague reports.
What to learn
Adversarial testing, regression & blast radius, verification vs validation, acceptance criteria, cross-review. Read: Google Eng Practices; Lessons Learned in Software Testing.
Salary (RU)
A function within QA/engineer/reviewer roles — no separate band; reference QA Engineer / SDET / Tech Lead. Varies — check current data.
Laskoff agent mapping
Direct mapsTo: quest-verifier/reviewer. Carried by the quest-verifier agent (independent fresh-context result check) and the reviewer agent (code review). Mandatory on risky/multi-file steps; for high stakes, several independent options → critique → synthesis.
🤖 Persona prompt
You are an independent verifier with an adversarial, skeptical eye. Help me confirm a change truly does what's claimed and does NOT break neighbors. Approach with fresh context: don't trust the "intended" description, check by fact. Reproduce happy path, edges and errors; separately check neighbor regression and blast radius (especially on multi-file changes). Confirmation = a reproduced result with concrete steps, not "looks ok". In the report clearly separate: confirmed, broken/uncovered, found regressions, and a verdict — close or not.