LONG-CONTEXT CODE REVIEW
Read wide. Review the change that matters.
A useful code review needs more than a diff. Give Space Bunny the architecture notes, relevant files, failing trace and change request so it can reason across the boundary instead of guessing from one snippet.
Review this change in the context of the repository.
1. Trace the request path across the affected files.
2. Identify correctness, migration and rollback risks.
3. Separate confirmed findings from assumptions.
4. Return file paths, evidence and a smallest safe next step.WHY CONTEXT MATTERS
The bug is often one boundary away from the diff.
Repository work becomes harder when the model sees only the changed function. Long context gives it room to compare interfaces, callers, tests, configuration and incident evidence in one working set.
Follow the architecture
Include module boundaries and the request path so the review can reason about how a local change travels through the system.
Keep evidence together
Combine source files, logs, tests and design notes when the failure only appears across more than one artifact.
Make uncertainty visible
Ask for evidence, assumptions and confidence separately instead of accepting a polished paragraph with no audit trail.
A PRACTICAL REVIEW LOOP
Give the model the whole question, not every file by default.
A million-token window is headroom. Start with a useful context packet, then expand only when the model identifies an information gap.
Map the change
List the changed files, entry point, expected behavior and the modules that own the relevant state.
Inspect the boundary
Add callers, interfaces, tests, configuration and logs that can confirm or disprove the suspected failure path.
Return an actionable review
Require file-level evidence, severity, assumptions and a smallest safe next step. Let people approve the change.
A REUSABLE REVIEW BRIEF
Ask for findings you can verify.
The prompt should constrain the review without pretending the model owns your repository or deployment process.
review = client.chat.completions.create(
model="stealth/space-bunny-alpha",
messages=[{
"role": "user",
"content": [
{"type": "text", "text": review_brief},
{"type": "text", "text": repository_context},
],
}],
extra_body={"reasoning": {"effort": "high"}},
)KEEP THE HUMAN IN THE LOOP
A broad read does not make the result automatically correct.
Space Bunny is an anonymous preview model. Exclude secrets, validate findings against the code, and keep writes, deployments and security decisions behind explicit application and human checks.
START WITH ONE REAL REPOSITORY QUESTION
Use the context window where the full picture changes the answer.
Try an incident review, cross-file refactor plan or migration risk check in the Playground, then move the same request behind your server.