Multi-model consultation for SDD specifications providing structured feedback. Coordinates parallel AI reviewers, synthesizes actionable insights, and categorizes findings by feedback type without modifying specs or executing fixes.
Inherits all available tools
Additional assets for this skill
This skill inherits all available tools. When active, it can use any tool Claude has access to.
reference.mdThe sdd-plan-review skill coordinates parallel AI reviewers to provide structured, categorized feedback on SDD specifications. It surfaces consensus findings, diverse perspectives, and prioritized recommendations to inform spec quality decisions.
Core Philosophy:
This skill is part of the Spec-Driven Development quality assurance family:
sdd-plan → sdd-plan-review (this skill) → sdd-modify → sdd-next → Implementation
Use this skill when you need to:
Do NOT use for:
sdd-fidelity-review)sdd-modify)This skill relies entirely on the Foundry MCP server (foundry-mcp). Tools use the router+action pattern: mcp__plugin_foundry_foundry-mcp__<router> with action="<action>".
| Router | Actions | Purpose |
|---|---|---|
review | spec-review, list-tools, list-plan-tools | Execute reviews and list toolchains |
spec | get, get-hierarchy, stats, list | Get spec content and structure |
provider | list, status, execute | List and manage AI providers |
Critical Rules:
Read() on spec JSON files directlycat, grep, jq for spec parsing| Type | Models | Duration | Focus | Use When |
|---|---|---|---|---|
| quick | 2 | 10-15 min | Completeness, Clarity | Simple specs, time-constrained |
| full | 3-4 | 20-30 min | All 6 dimensions | Complex specs, moderate-high risk |
| security | 2-3 | 15-20 min | Risk Management | Auth, data handling, compliance |
| feasibility | 2-3 | 10-15 min | Estimates, Dependencies | Tight deadlines, uncertain scope |
For detailed dimension descriptions, see
reference.md#review-dimensions
Run a review:
mcp__plugin_foundry_foundry-mcp__review action="spec-review" spec_id="{spec-id}" review_type="{type}"
Check available tools:
mcp__plugin_foundry_foundry-mcp__review action="list-tools"
Check provider status:
mcp__plugin_foundry_foundry-mcp__provider action="list"
mcp__plugin_foundry_foundry-mcp__review action="list-tools"
Check which AI review toolchains are installed:
Use this decision matrix:
| Condition | Review Type |
|---|---|
| Task count ≤ 10 AND no auth/payment/PII tasks | quick |
| Task count > 10 OR has architectural decisions | full |
| Contains auth, payment, encryption, or PII handling | security |
| Phase has deadline pressure OR external dependencies | feasibility |
Priority when multiple conditions match:
security (always takes precedence for sensitive data)full (for complex specs)feasibility (for time-constrained work)quick (default fallback)mcp__plugin_foundry_foundry-mcp__review action="spec-review" spec_id="{spec-id}" review_type="{type}"
The tool automatically:
Priority Levels:
For consensus interpretation details, see
reference.md#interpreting-consensus
After completing the review:
Reports are written to specs/.plan-reviews/{spec-id}-review-{type}.md:
## Feedback Summary
### Risk Flags (CRITICAL)
1. Missing authentication on admin endpoints
Priority: CRITICAL | Flagged by: gemini, codex
Impact: Unauthorized access to sensitive operations
Recommendation: Add JWT validation middleware
For complete output template, see
reference.md#output-format
Why Isolation Matters: Review output can be extensive (hundreds of lines). Writing directly to the conversation wastes tokens.
Isolation Behavior:
Output Locations:
specs/.plan-reviews/{plan-name}-{review-type}.mdSkill(foundry:sdd-plan-review) "Review spec user-auth-001 with 'full' review type. Check all 6 dimensions and surface consensus findings."
| Parameter | Value |
|---|---|
| Min models required | 1 |
| Recommended models | 2+ |
| Timeout per model | 120s |
| Max review time | 300s |
For full configuration, see
reference.md#configuration
For comprehensive documentation including:
See reference.md