Skill — code-review
Nishad
All Companies ▸ FEATURE-1755 — Task Detail ▸ code-review
← Back to task detail
Skill / step definition — two-tier (mirrors om-forge step-detail, TECH-668 §3/H-9)

Step: code-review

FEATURE-1755 · step 5 of 9

Reached from the task drill: FEATURE-1755's workflow runs this step. Read-only view below shows what it does, in plain English; a SuperUser can edit the skill body / step config directly, further down.

What this step does — read only

Plain English, no skill filenames, role pools, or model ids. This is everything anyone without SuperUser access can see about this step.

Input
The committed code, and the plan it's meant to match
Process
A different reviewer checks the code for correctness, security, and completeness against the plan
Who
A reviewer — never the person who wrote the code
Output
A code review report, with pass/fail on each check
💬
Ask Cortex about this step
Plain English, top-right — "skip this for small copy changes," "add a second reviewer." 🎤 or type · 🔊 Cortex replies aloud too

Direct field-editing (below) exists only at the SuperUser tier.

SuperUser editing
⚠️ Editing this step makes it a customized workflow — you own its correctness; platform updates stop auto-applying to customized steps.
● customized · Cortex-reviewed

Shown for illustration — this step is not currently customized. When a SuperUser edits any field below, the badge above updates automatically after Cortex's review; e.g. flagged:

● customized · Cortex-flagged
Finding: removing the "Logic / security" check leaves no automated check for injection vulnerabilities on this step — advisory only, not blocking; the customization still applies as edited.

Core properties

Skill

Edit

This step runs dev-review — the SOP that defines what "reviewed" means for a committed code diff. Every agent assigned this step loads the same skill; the skill, not the agent, is the unit of behavior.

Model / skill pairing

Edit

Opus — the review + judgment tier.

Pairing = which model tier is allowed to run this skill. Per the model-routing rubric: review / judgment → Opus · test / deploy → Sonnet · research / plan → Gemini · gates → human.

Artifacts dependsOn / artifactOut

Edit
In
The committed code diff · PLAN-nn.md · REQ-nn.md
Out
A code review report, appended to the task

Validation checks evidenceRequired

Edit

The gate checklist this skill enforces, PASS/FAIL each.

  • Plan-completeness
    Diff covers every step in PLAN-nn.md
    PASS
  • Logic / security
    No correctness or vulnerability findings
    PASS
  • No-silent-failures
    Guardrail #12 — no swallowed errors
    PASS
  • Runtime-completeness
    Rule 1 — backing migration + wired path exist
    PASS

Status ladder position

  • coding
  • →
  • code-review
  • →
  • testing
↩ on fail, sends back to rework

★ Additions — confirmed extras

Gate ★ addition

Edit

Cleared by a reviewer agent — not a human. producer ≠ reviewer is enforced at task_lock: the coder who wrote the diff can never review its own code. resolveApprover picks an eligible reviewer at run time.

Eligibility / role pool ★ addition

Edit

Which agents / models are eligible to run this step: project_role_pools today → the Capability Ledger tomorrow.

Per-step cost / telemetry ★ addition

Every external model call this step makes writes a usage record.

Last run: 1 Opus call · $0.04 · 2,150 tok

What's on this screen

Built from: the written procedure assigned to this workflow step, this task's status ladder, and Cortex's own review of anything you customize.

On screenWhere it comes fromWhat it means to you
Step position (e.g. "step 5 of 9")where this step sits in the task's overall workflowknow how many steps remain before the task is done
Input · Process · Who · Outputa plain-English description of what this step does, pulled from its procedureunderstand what happens here without needing edit access
Ask Cortex about this stepa plain-language question box scoped to this one stepget a direct answer instead of hunting through docs
Customization warninga heads-up that shows once you start editing this steptells you that from here on, you're responsible for keeping it correct, not the platform
Cortex-review badge and findingCortex's own check of any edit you make to this stepflags a gap in your edit before you rely on it
Skill (e.g. "dev-review")the actual step-by-step procedure every agent follows when running this stepsee exactly which procedure defines "done" here
Model / skill pairingthe rule for which tier of model is trusted to run this kind of stepthe quality bar guaranteed for this step
Artifacts — In / Outwhat this step reads before it starts and what it hands off when it finishestrace what feeds this step and what depends on it next
Validation checks (pass/fail checklist)the specific checks this step must clear before it can passsee exactly what was verified, not just a pass/fail headline
Status ladder positionthis step's place between the step before and after it, including where it goes on failureknow what happens next if this step passes or fails
Gatewho has to clear this step — always someone other than whoever did the workconfirms no one reviews their own work
Eligibilitywhich agents are currently allowed to be assigned this stepknow who's eligible if you need to reassign it
Per-step costwhat the last run of this step actually cost in model spendkeep an eye on what each step costs to run