A bounded problem
Supply expected versus actual behavior, reproduction steps and the smallest relevant code excerpt. State what must remain unchanged so a model does not turn a narrow fix into a rewrite.
Engineering teams, founders, and technical operators
Turn a reproducible bug or bounded feature request into a proposed change, an explicit test plan and a reviewable handoff. Use AI chat to reason about supplied code; apply and test changes in your own development environment.
Practical guidance
A small change proposal containing the problem, evidence, affected files, acceptance tests and remaining risks—not an automatically merged or deployed fix.
Supply expected versus actual behavior, reproduction steps and the smallest relevant code excerpt. State what must remain unchanged so a model does not turn a narrow fix into a rewrite.
Include language/framework versions, relevant types and sanitized error output. Label each file or excerpt; explain omitted context. Never paste credentials, access tokens, customer records or confidential code without authorization.
Write the tests or observable outcomes that would establish success. Include one normal case, one edge case and one failure case. Keep a baseline result from your local environment for comparison.
Practical guidance
Ask for a concise problem statement, likely causes and missing evidence using only the supplied material. Have the model distinguish observations from hypotheses.
Checkpoint: Every claimed observation should point to an excerpt or reproduction step you provided.
Request two approaches with affected files and tradeoffs. Choose one based on compatibility, review size and testability, not the confidence of the answer.
Checkpoint: Reject unrelated rewrites, invented APIs and assumptions about files the model has not seen.
Request a minimal diff or clearly labeled replacement block plus a regression test that would fail before the change. Ask for environment-specific commands separately.
Checkpoint: Check imports and installed versions yourself before applying any suggested command or dependency.
Apply the proposal in your own branch or sandbox. Run the relevant tests, lint and type checks. Return the actual failure output to AI chat for a focused revision if needed.
Checkpoint: An AI statement that tests pass is not evidence. Record the command and its observed result.
Summarize why the change works, what was tested and what remains uncertain. Review the complete diff before merging through your normal development process.
Checkpoint: Include security, accessibility or compatibility concerns when they are relevant to the changed behavior.
Technical details
Workflow boundary
This workflow uses material you supply to AI chat. It does not establish repository synchronization, shell execution, automated pull requests or deployment from EZ Ai Assist. File access, attachments and tool availability must be checked in your actual session.
Editorial practice
Identify exact versions and request a diff, test file or review table explicitly. Mark unknown functions as unknown. Break a large problem into independent changes; model context size alone does not guarantee every detail will be retained.
TypeScript documentation + workflow guidance
TypeScript documents narrowing through checks such as typeof, equality and control flow. For untrusted input, ask for explicit validation before relying on a type assertion; verify the relevant branches with runtime tests in your own environment.
External safety reference; not an app integration
GitHub's Copilot code-review guidance warns that suggestions may be inaccurate or insecure and calls for human review and testing. That is a useful review principle here, not a claim that EZ Ai Assist includes Copilot or its repository integrations.
Technical details
These are editorial starting points, not measured comparisons. Check the exact model guide and current app access before using a model; linked catalogs may also contain historical entries.
Evaluate a reasoning-oriented chat model on a bounded failure with several plausible causes. Ask it to explain which supplied evidence separates those causes.
Tradeoff: More reasoning cannot supply missing runtime state or replace reproduction.
Explore Anthropic modelsReuse the same brief and acceptance criteria with another available model family when the proposed change is risky or unclear.
Tradeoff: Agreement between models is not independent proof that a fix works; test both against the same cases.
Compare Mistral AI modelsUse the exact model guide to check documented input limits before evaluating a larger code excerpt. Supply a dependency map and only the relevant surrounding code.
Tradeoff: A provider context ceiling does not establish the app's upload limit, retrieval behavior or repository access.
Read the Kimi provider guidePractical guidance
Replace every bracketed placeholder with authorized context. These are original templates, not generated-output demonstrations or generation-tested recipes. Copying does not submit anything or incur a generation charge.
Where to use it: Paste into AI Chat with sanitized code and context.
Investigate [EXPECTED BEHAVIOR] versus [ACTUAL BEHAVIOR] in [LANGUAGE/FRAMEWORK AND VERSION]. Reproduction: [STEPS]. Relevant code and error output: [LABELED EXCERPTS]. Preserve [CONSTRAINTS]. Separate observed evidence from hypotheses. Give the smallest proposed change, affected files, missing information and a regression test. Do not claim to run code or inspect files I have not supplied.
Check: Confirm the proposal addresses your reproduction and cites the correct excerpt.
Where to use it: Paste into AI Chat after running the proposal locally.
Revise this proposed change: [DIFF]. I ran [EXACT COMMAND] in [ENVIRONMENT] and observed [SANITIZED OUTPUT]. Acceptance criteria: [CRITERIA]. Explain the failure using this evidence, then propose the smallest correction and an additional regression case. Preserve working behavior. If context is missing, ask for it instead of inventing dependencies or reporting an unrun test as passed.
Check: Run the revised test yourself and compare the changed diff with the original scope.
Where to use it: Paste into AI Chat with the actual diff and test record.
Review [FINAL DIFF] against [ACCEPTANCE CRITERIA]. Observed checks: [COMMANDS AND RESULTS]. List concrete findings with file or excerpt references, impact and suggested verification. Separate confirmed defects from possible risks. Check edge cases, data handling and compatibility. Finish with a concise handoff that distinguishes passed checks, failed checks and checks not run. Do not approve deployment on my behalf.
Check: A clean review is not a guarantee; inspect the final diff and retain your normal approval process.
Illustrative example
An editorial example of structure, not a generated result, customer case study or measured outcome.
Example: a blank search submits an unnecessary request. Expected behavior: preserve existing results and do not submit whitespace-only queries. Excluded: changing the backend or redesigning the search UI.
Tests to write: blank query, whitespace query, valid query and existing error handling. Proposed commands and results are separate fields; leave results as Not run until execution supplies evidence.
Attach the reviewed diff, observed results and outstanding risks. A teammate should be able to reproduce the original issue and verify the change without relying on the chat transcript.
Technical details
Provide the installed version and current official documentation excerpt. Ask for an implementation compatible with that version, then verify the import and signature locally.
Restate the minimal affected area, supply the current diff and explicitly exclude unrelated refactors. Ask for a patch against this version only.
Ask it to relabel proposed tests as unrun. Only execution output from your own environment establishes a test result.
Provider maximums, picker estimates and actual customer charges are different things. Read the pricing explanation.
Common questions
No repository or deployment integration is established by this guide. Copy the relevant, authorized context into chat, then apply and verify suggestions using your own tools.
There is no universal winner established here. Evaluate available models against a small representative task and the same tests. The linked guides explain model-specific capabilities and limitations, not benchmark results from this workflow.
Use the smallest relevant sanitized excerpt. Remove credentials, personal data and confidential content, and follow your organization's data-sharing rules before submitting code or logs.
Evidence and review
The linked technical references were reviewed on . Documentation and app availability can change.
Workflows and templates are original editorial guidance, not benchmarked model runs. Official references support the attributed technical points; they do not establish app tools, repository access or successful execution. Linked model/provider guides retain their own evidence dates.
Opening EZ Ai Assist does not prefill these templates or automatically run the workflow. Choose the relevant mode and review your input before submitting it.
Continue exploring
Create SOPs, process maps, internal guides, training content, meeting summaries, and implementation checklists.
Turn customer context, research, and roadmap ideas into PRDs, user stories, launch notes, and decision briefs.
Research markets, compare competitors, summarize public information, and turn findings into decision-ready briefs.