Where are you stuck?
You do not need a polished answer. Enter what you heard and what you know.
Your entries stay in this browser and are never sent anywhere by this page.
- 1. Where are you stuck?
- 2. Write down the work you actually received
- 3. Can you tell whether the result is correct?
- 4. What happens if the result is wrong?
- 5. Where are the materials and the work?
You do not need a polished answer. Enter what you heard and what you know.
Current policy
What this task appears to be
What remains unknown
What the AI may do
What the AI must not do
Evidence required before trust
Questions for the requester
AI work handoff
AI work handoff
Check the result you received from the AI
You do not need to paste the answer. Compare the real result against each gate below.
Principles behind the tool
Open the operational concepts hidden from the beginner flow only when needed.
Decision order
- Clarify first when the request is ambiguous and the requester is reachable.
- Add unfamiliar-domain investigation and human review when you cannot judge the result.
- Block external actions and final decisions when impact is high or unknown.
- Choose one current stage: investigation, review, integration, or production.
- Default to sequential work when ambiguity or dependency exists.
- Choose where the work and evidence live before choosing a product.
- Do not call work done until inspectable evidence passes.
Parallel implementation is never auto-selected. Use it only when file boundaries, runtime isolation, independent checks, and an integration owner are all proven.
Operating modes
Clarify the requestOffer interpretations and questions without choosing the requester's intent.
Create request interpretations and confirmation questions before producing the artifact.
clarify
Frame the first stepDefine the goal, constraints, and smallest checkable first step.
Define a 30-minute first step and the condition that proves it is finished.
thinking
Investigate evidenceCollect inspectable sources, options, and unknowns before a conclusion.
Start with a research note that attaches a source and date to important claims.
investigation
Map an unfamiliar domainExpose the domain, examples, assumptions, and human judgment points before asking for an answer.
Ask the AI for a domain map and review points, not a final answer.
unknown
Produce a reviewable artifactCreate a bounded draft and show assumptions and verification steps.
Ask for a reviewable draft and exact checks instead of a final submission.
implementation
Independent reviewFind defects, omissions, and risks with evidence without assuming the artifact is correct.
Request defect locations, severity, and evidence before any rewrite.
review
Integrate with conflicts visibleCombine results while preserving contradictions, resolutions, and open decisions.
Require a conflict log and an end-to-end check beside the integrated result.
integration
VerifyJudge the result by reproducible evidence, not the AI's confidence.
Judge the result by reproducible evidence, not the AI's confidence.
verify
Targeted revisionFix only the failed proof gate and run the same check again.
Fix only the failed proof gate and run the same check again.
revise
Capture a reusable ruleSave only routes and checks that proved useful repeatedly.
Save only routes and checks that proved useful repeatedly.
capture
Sequential workVerify evidence from one stage before starting the next.
Verify evidence from one stage before starting the next.
sequential
Parallel workUseful for independent questions; simultaneous editing requires proven boundaries and integration ownership.
Useful for independent questions; simultaneous editing requires proven boundaries and integration ownership.
parallel
Capabilities and product examples
These are capability routes, not product rankings. Features and availability can change, so verification dates and official sources are shown.
Conversational framing and drafting
Clarification, structure, writing, and critique
Product examples verified 2026-07-21ChatGPT or Claude
Source-based web research
Multi-source research with traceable citations
Product examples verified 2026-07-21ChatGPT Deep Research or a source-restricted research agent
Local file and repository work
Read and edit files, then run real verification
Product examples verified 2026-07-21Codex, Claude Code, Gemini CLI, or GitHub Copilot CLI
Browser and connected-app work
Inspect signed-in systems and perform bounded actions
Product examples verified 2026-07-21A browser or connected-app agent with confirmation gates
Product information sources
- OpenAI Codex appCreated 2026-02-02; Windows update 2026-03-04
- Codex for every role and workflowCreated 2026-06-02
- OpenAI deep researchCreated 2025-02-02; updated 2026-02-10
- Gemini CLI documentationLast updated 2026-06-18 when verified
- Claude Code subagentsNo visible creation/update date when verified 2026-07-21
- GitHub Copilot coding agentNo visible creation/update date when verified 2026-07-21