mdmagnuson-creator/critic-dispatch

Critic Dispatch Skill

Control when @critic runs during PRD work to balance thoroughness vs speed.

Triggers

• Starting PRD execution
• After completing a story (to check if critic should run)
• "critic mode", "when to review", "code review batching"

Applicable Agents

• builder — orchestrating PRD execution and story completion
• developer — when self-reviewing after significant changes

---

Configuration Cascade

Resolved in order (highest priority first):

1. CLI flag — --critic-mode=strict (one-off override)
2. Project-level — project.json → agents.criticMode
3. Hardcoded fallback — balanced

---

Critic Modes

| Mode | Behavior | Use Case | |------|----------|----------| | strict | Run @critic after every story | High-risk projects (payments, auth, security) | | balanced | Run @critic every 2-3 stories | Default — catches issues without excessive overhead | | fast | Run @critic once at end of PRD | Greenfield projects, low-risk changes, speed priority |

---

Balanced Mode Logic

• Run critic after story 2, then every 3 stories (story 5, 8, 11, etc.)
• If PRD has ≤2 stories, behave like fast (one critic run at end)
• Always run critic at PRD completion regardless of mode

Formula

shouldRunCritic = (storyNumber == 2) || (storyNumber > 2 && (storyNumber - 2) % 3 == 0)

Examples

| Story # | Run Critic? | Reason | |---------|-------------|--------| | 1 | No | First story, wait for context | | 2 | Yes | First checkpoint | | 3 | No | Too soon after 2 | | 4 | No | Still building | | 5 | Yes | 3 stories since last review | | 6 | No | Continue | | 7 | No | Continue | | 8 | Yes | 3 stories since last review |

---

Implementation

At PRD Start

1. Determine critic mode from config cascade
2. Log: "Critic mode: [mode]"

After Each Story Completes

if criticMode == "strict":
    run @critic
elif criticMode == "balanced":
    if storyNumber == 2 or (storyNumber > 2 and (storyNumber - 2) % 3 == 0):
        run @critic
elif criticMode == "fast":
    # Skip until PRD completion

At PRD Completion (All Modes)

run @critic  # Final review before PR

---

Critic Invocation

When running @critic:

1. Pass relevant context:

   • Files changed since last review
   • Story descriptions completed
   • Test results summary

2. Wait for completion:

   • Address blocking issues before continuing
   • Log advisory issues for later

3. Track in chunk.json:

   {
     "lastCriticRun": {
       "afterStory": 2,
       "timestamp": "2026-02-28T10:00:00Z",
       "blockers": 0,
       "advisories": 3
     }
   }
   

---

Override Examples

CLI Override (One-Off)

# Run critic after every story for this session
builder --critic-mode=strict

# Skip intermediate reviews, only final
builder --critic-mode=fast

Project-Level Configuration

// docs/project.json
{
  "agents": {
    "criticMode": "strict"
  }
}