/wbReview — Exhaustive Simulation ()
/wbReview is the inspector. Its job is the plan-vs-reality diff: read what was supposed to happen (the plan), read what actually happened (the diff or the current state), and surface where they disagree. Unlike /wbAudit, which judges code against universal standards, /wbReview judges code against intent. Same code can pass an audit and fail a review — or vice versa — because the questions are different.
Read this if you want to know what --plan does that /wbAudit --wbPlan doesn't, why --act and --wbPlan are independent flags here too, and where review ends and refactor begins.
1. Role & target
| Aspect | Behavior |
|---|---|
| Role | The Inspector — surgical plan-vs-reality check. |
| Target | An uncommitted diff, a recent commit, a directory, or a plan row. |
| Cell scope | None directly. Review produces a report; chaining (--act / --wbPlan) routes the report into actionable artifacts. |
| Side effects allowed | Reading code, reading the active plan, comparing diff to plan row descriptions, reading recent reports under reports/<date>/. |
| Side effects forbidden | Editing source code, applying suggestions, mutating plan cells directly. |
The plan-vs-reality framing is the design center. Three concrete asymmetries follow from it:
- Scope creep is reviewable. If a plan row asks for "fix regex escape" and the diff also includes a CSS class change,
/wbReviewflags it./wbAuditwould not — extra CSS isn't audit-worthy on its own. - Missing work is reviewable. If the plan row says "add denylist + tests" and the diff includes the denylist but no tests,
/wbReviewflags it./wbAuditlooks at what's there, not what's missing. - Style consistency with the rest of the codebase is reviewable. If the diff introduces a pattern alien to the existing code,
/wbReviewsurfaces the divergence even when the new pattern is "fine on its own."
These three together give review its distinct lane.
2. Argument resolution matrix
| Form | Example | What /wbReview does |
|---|---|---|
| No argument | Command: /wbReview | Reviews uncommitted changes (git status + diff). Maps each touched file back to the most-recent plan row that owns it. |
| Plan row ID | Command: /wbReview --plan (alone) | Equivalent: review uncommitted changes through the plan lens. |
| Specific commit | Command: /wbReview <sha> | Reviews a single commit. Maps to plan rows referenced in the commit body. |
| Directory | Command: /wbReview core2/packages/wb-core/ | Reviews all files in the directory against the plan rows that touch them. |
| Free-text scope | Command: /wbReview "the auth changes" | Refused. Same reasoning as /wbAudit — adversarial review is too sensitive to fuzzy targeting. |
Key behavior: when reviewing a diff, the agent resolves "which plan row owns this change" by reading recent reports/<date>/plans/plan_*.md files and matching files-touched against the row's description. If no plan row claims a touched file, that file is reviewed under "scope-creep" by default.
3. Flag matrix
| Flag | Shortcut | Mode | What it does |
|---|---|---|---|
--plan | -p | Filter | Strict plan-conformance mode. Failures on any deviation (scope creep, missing required artifact, style divergence). |
--act | -a | Chain | Routes review output through /wbActOn — produces a sibling action file ranking review findings by impact/effort. |
--wbPlan | -P | Chain | Routes review output through /wbActOn's plan-generation engine — appends new rows to the active plan, one per critical finding. |
--act and --wbPlan are independent and composable, same shape as /wbAudit.
How --plan changes the verdict
Without --plan, /wbReview is advisory: it flags issues but doesn't refuse. With --plan, it gates: any finding becomes a hard "review failed" verdict.
| Mode | Scope creep finding | Missing artifact | Style divergence |
|---|---|---|---|
| Default (advisory) | Flagged with severity | Flagged with severity | Flagged with severity |
--plan (strict) | Hard fail | Hard fail | Hard fail |
The advisory/strict split exists because reviews happen at different points in the workflow. Mid-task, advisory is more useful (you want to know without being blocked). Pre-merge or pre-release, strict is what you want.
4. Pipelines (the agent-native scenarios)
/wbReview> /wbReview💠 Pipeline The mid-session "am I drifting?" check
A /wbWork session has been running for an hour. The user wants a quick advisory pass: did anything I just did stray from what the plan asked for?
💠 Pipeline The strict pre-validation gate
Same situation, but the user is about to run /wbValid and wants to be sure nothing slips:
💠 Pipeline The chained "review → plan rows for the gaps"
A long session has accumulated review findings that warrant follow-up work. Instead of fixing in place, route the findings into new plan rows:
5. Edge cases & refusals
| Trigger | What /wbReview does |
|---|---|
| No diff, no commit, no target | Halt. ❌ Nothing to review — no uncommitted changes, no commit specified. |
| Free-text target | Refused with disambiguation list. |
| Target file genuinely matches no plan row | Reviewed under "scope-creep." Not a refusal — the user might be fixing typos that don't need a row. |
Diff includes only generated files (dist-*, lock files) | One-line notice. Generated diffs aren't review-worthy on their own. |
/wbReview --plan with no active plan | Halt. Strict mode requires a plan to compare against. |
--plan --wbPlan with no findings | Produces an empty review report; no plan rows added. The chained mutation is conditional on findings. |
| Target commit references a plan row that no longer exists in the active plan | Notice that the row is missing; reviews against commit body's description verbatim instead. |
The pattern: /wbReview is a plan-aware critic. It reads intent (the plan) and reality (the diff), names where they disagree, and either advises or gates depending on --plan. The chaining flags turn findings into either ranked actions (--act) or new plan rows (--wbPlan), so review failures become tracked work rather than friction. The unifying principle: a review without intent is just an opinion; the plan is what makes it adjudicable.