Install Selective Intelligence, choose a failure you already have, and judge it against a visible pass/fail line. No invented success rate and no telemetry.
What installation does not do: it does not grant permission to publish, spend, delete, deploy, disclose, or change access.
Use work you can safely share with your AI. Remove secrets and personal or proprietary information before public feedback.
Use this when the AI produced something polished but fundamentally wrong.
Selective Intelligence: that is not what I meant. Re-read my request and your result, identify the misunderstanding, fix the real cause, and verify the corrected outcome.
Pass: it changes the work, preserves your correction, and checks the corrected result. Fail: it defends the old answer, asks you to repeat everything, or only apologizes.
Use this when you have a goal but not a developer-ready specification.
Selective Intelligence: turn this rough idea into the complete result I actually need. Make safe reversible choices, finish the usable deliverable, and verify it before calling it done.
Pass: you get the real deliverable with important unknowns visible. Fail: you get only an outline, questionnaire, template shell, or invented facts.
Use this on an unfinished repository, document, workflow, or long-running chat.
Selective Intelligence: audit this unfinished project, recover the intended outcome, reuse what works, finish what is missing or disconnected, and prove the main user journey works.
Pass: it maps intended versus reachable work and continues from the current state. Fail: it starts over, counts files as progress, or treats a passing build as the finished outcome.
Use this when the AI keeps rereading, replanning, or generating duplicates.
Selective Intelligence: complete this with a lean context. Read only what changes the decision, reuse existing work, avoid repeated plans and filler, and keep the required verification.
Pass: less irrelevant context and repetition without shrinking the outcome. Fail: shorter output that drops required work, evidence, or verification.
Use this when the AI says done but you cannot see proof.
Selective Intelligence: do not assume this is complete. Inspect the actual result, trace every requested outcome to current evidence, fix the highest-impact gap, and state exactly what is verified, unverified, or blocked.
Pass: completion is tied to current observable evidence. Fail: it cites intention, code presence, a plan, or an old test as proof of the real result.
After the task, report Worked, Partly, or Wrong. A short redacted description is enough. Public feedback is optional; the plugin does not collect prompts or project contents automatically.