Panda Review
Examples use standalone Codex names. For a Codex plugin, use $panda-suite:panda-review; for a Claude Code plugin, use /panda-suite:panda-review. See all invocation formats.
Panda Review finds actionable defects introduced or exposed by a change: a pull request, a diff, or another bounded comparison. It checks correctness, security, reliability, data integrity, compatibility, and regression risk, each with precise evidence.
At a glance
| Topic | Summary |
|---|---|
| Use it for | PR review, reviewing your own branch before merge, checking a diff against a requirement |
| Not for | Complexity advice without a defect (Simplify), a dedicated vulnerability or dependency audit (Security) |
| Edits by default | None. Review does not authorize fixes |
| Main output | Prioritized findings, or a short "no significant findings" result with review limits |
| Never does | Invent findings to fill a report, or present speculation as a confirmed vulnerability |
How it works
- Understand intent. Requirements and repository guidance, the diff, and the affected callers, contracts, state transitions, and tests.
- Trace behavior. Validation, authentication and authorization, errors, transactions, concurrency, data integrity, API compatibility, and operations where relevant. Expected behavior comes from the contract before existing tests are trusted: passing tests alone do not prove correctness.
- Validate candidates. The triggering input or condition is traced through real code and existing guards, with a focused safe reproduction when useful. Findings already prevented by the code, or based only on preference, are dropped.
- Report by impact. Material defects before style, with precise file and line locations.
Finding format
| Field | Content |
|---|---|
| Severity | critical (immediate severe compromise or data loss), high (core correctness or security), medium (a bounded path), low (limited impact) |
| Location and issue | File/line and the faulty behavior |
| Trigger and evidence | The input or state that triggers it and the causal path, including mitigations considered |
| Impact and correction | The consequence and the smallest appropriate repair |
For example: removing an owner check lets a non-owner cancel an unshipped order when no caller guard exists. The finding cites the changed function and the caller contract, not just "missing security".
Examples
Branch against a requirement
$panda-review Review the current branch against origin/main for the owner-only
cancellation requirement. Trace changed behavior and callers. Report material
defects with severity, precise location, trigger, evidence, impact and correction.
Do not fix code.
Current uncommitted diff
$panda-review Review the current diff for material defects.
Explain the trigger, impact, and location of each finding. Do not fix code.
A pull request
$panda-review Review PR 128 (branch feature/bulk-import against main). Focus on
data integrity during partial failures.
A specific commit range
$panda-review Review commits a1b2c3d..HEAD. These should be a pure refactor;
flag any observable behavior change.
Compatibility focus
$panda-review Review the API changes in this branch for backward compatibility
with existing mobile clients. Use docs/api.md as the contract.
Concurrency and transactions
$panda-review Review the new payment retry logic for double-charge risks under
concurrent requests and process restarts.
Review, then fix
$panda-review Review the diff, then fix the high and critical findings only and
re-run the tests.
Fixes need explicit authorization like this; review alone never edits code.
Under pressure to approve
$panda-review We need to merge this in ten minutes. Is it safe?
Time pressure does not change the evidence. Expect the relevant guards inspected and supported findings reported, or a clear "no significant findings" with limits.
Boundaries
Application files stay unchanged unless fixes are requested. External or destructive verification needs authorization. A small diff review never implies a whole-system audit; the reviewed range is always stated.
Related
Skill instructions
Read the full Panda Review instructions in the repository.