Design review gate

Skill zekiyugnak/telar-framework/skills/orchestration/design-review-gate

Install

npx -y skills add zekiyugnak/telar-framework --skill design-review-gate

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

6.1 KB, as published. Nobody here has run it

Design Review Gate

Trigger condition (binding)

This skill is loaded only via:

The orchestrator agent's workflow after a RESEARCH.md or design doc commit.
/tl-telar:review-design [--latest | --design-file <path>] (sets TL_TELAR_ORCHESTRATED=1).
skills/brainstorm-first.md's Orchestrated Mode section (sub-spec 6 augmentation, §1.1-gated).
Explicit user request such as "run the design review gate".

This skill is NEVER auto-triggered from legacy mobile commands. /tl-telar:create-app, /tl-telar:add-feature continue to use brainstorm-first → plan-and-track flow WITHOUT the design gate.

What this skill does

For a given design doc (typically RESEARCH.md or docs/plans/*-design.md):

Spawn 6 fresh Task() reviewers in parallel (one tool-call batch via Promise.all-shape): PM, Architect, Designer, Security-Design, CTO, Mobile-Platform.
Aggregate per-reviewer ReviewResult JSON.
Compute overall verdict: APPROVED iff all 6 returned APPROVED; NEEDS_REVISION otherwise.
On APPROVED: hand control back to orchestrator (or user, if standalone). Emit a structured summary.
On NEEDS_REVISION (iteration < 3): present consolidated blockers/suggestions/questions to user. User revises design. Re-run all 6 FRESH (iteration + 1). Max 3 iterations.
On 3rd-iteration NEEDS_REVISION: escalation report with Override / Defer / Cancel.

Step-by-step procedure

Step 0: Resolve inputs

Parse --design-file <path> or --latest (most recent design doc in docs/plans/*-design.md or RESEARCH.md at repo root).
Read design doc text. If empty or stub: abort.
Resolve userRequest from conversation context or prompt user.
Set iteration = 1.

Step 1: Spawn 6 reviewers in parallel

Requires a top-level caller — /tl-telar:review-design or the main-session orchestrator (see agents/orchestrator.md → "Execution context"). A Claude Code subagent has no Task tool; never let a spawned subagent invoke this skill.

Single Task() batch call. Each:

subagent_type: general-purpose
description: Design review — <role>
prompt: template from references/reviewer-prompts.md with {{userRequest}}, {{designDoc}}, {{iteration}} substituted.

Step 2: Collect verdicts

Parse 6 JSON responses. Malformed → re-prompt that single reviewer once (fresh Task()). Second failure → synthetic NEEDS_REVISION with rule X1-malformed-response.

Step 3: Aggregate

{
  "overall_verdict": "<APPROVED if all 6 APPROVED, else NEEDS_REVISION>",
  "iteration": <current>,
  "blocking_reviewers": [<roles with NEEDS_REVISION>],
  "all_blockers": [<concatenated>],
  "all_suggestions": [<concatenated>],
  "all_questions": [<concatenated>],
  "use_case_analysis": <from PM reviewer if present>,
  "threat_model": <from Security-Design reviewer if present>,
  "max_iterations_reached": <iteration >= 3>
}

Step 4a: APPROVED

## Design Review Gate — APPROVED (iteration N)

All 6 reviewers approved. Use case + threat model summarized below.

### Use case (from PM)
- WHO: {{...}}
- WANTS: {{...}}
- SO THAT: {{...}}
- WHEN: {{...}}

### Threat model (from Security-Design)
- High risk: {{...}}
- Medium risk: {{...}}
- Mitigations required: {{...}}

### Suggestions (non-blocking; for the planner to consider)
- [<role>] {{...}}

Hand back to orchestrator or end (standalone).

Step 4b: NEEDS_REVISION, iteration < 3

## Design Review Gate — NEEDS_REVISION (iteration N of 3)

Blocking reviewers: <list>

### Blockers (must address before next iteration)
1. [<role>] <rule>: <summary>
   - Evidence: <file>:<line>
   - Explanation: <...>

### Suggestions (consider, but not blocking)
...

### Open questions
...

### Next step
Revise the design doc addressing each blocker. Re-run /tl-telar:review-design.

Return aggregated verdict. Do NOT auto-loop.

Step 4c: NEEDS_REVISION, iteration ≥ 3 → escalation

## Design Review Gate — ESCALATION (3 iterations exhausted)

Iteration history:
| Iter | Blocking reviewers | Blocker count |
|------|---------------------|---------------|
| 1    | ...                 | N             |
| 2    | ...                 | N             |
| 3    | ...                 | N             |

### Decision required
1. **Override** — proceed despite findings.
2. **Defer** — set this design aside and revisit later.
3. **Cancel** — abandon the feature.

Choose 1/2/3.

Anti-patterns

Reusing reviewer Task() instances. Same as all gates — fresh every time.
Inlining prior reviewer findings into a new spawn. Each spawn sees only doc + request.
Auto-approving on partial APPROVED. All 6 must approve.
Treating Security-Design or PM as optional. Both are MANDATORY reviewers — their structured output (threat_model, use_case_analysis) is consumed downstream.
Skipping the gate for "simple" designs. If a brainstorming session committed RESEARCH.md and you're in orchestrated mode, the gate fires. Bypass requires explicit user override.

Tests / conformance

Run node scripts/validate-skills.js (orchestration-namespace checks). Run node scripts/validate-rubrics.js (rubric exists and is well-formed).