arch-review

Note: This skill intentionally saves to reviews/phase-2-architecture/ because it re-validates the Phase 2 architecture after RTL implementation. It updates (not replaces) the existing Phase 2 review artifacts with post-RTL findings per Hierarchical Spec Compliance.

<Use_When>

• RTL implementation is complete and needs architecture sign-off
• Microarchitecture spec (docs/phase-3-uarch/*.md) exists for comparison
• Pre-tapeout architecture review gate
• Suspecting architectural mismatch after late RTL changes </Use_When>

<Do_Not_Use_When>

• RTL is not yet lint-clean (fix lint first with rtl-lint-check)
• Only functional bugs need diagnosis (use rtl-bug-repro instead)
• Architecture changes are also requested (arch-review is READ-ONLY) </Do_Not_Use_When>

<Why_This_Exists> Architecture review catches structural mismatches between spec and implementation that functional tests cannot detect. Three independent reviewers with different focus areas (structure, timing, quality) provide broader coverage than a single pass. </Why_This_Exists>

<Execution_Policy>

• All three agents run in parallel, READ-ONLY
• rtl-architect checks spec-to-RTL consistency
• timing-advisor checks pipeline depth, clock domain structure, timing feasibility
• rtl-critic checks code quality, maintainability, synthesis-friendliness
• Findings aggregated into single report — no RTL changes </Execution_Policy>

 ## Feature Coverage Checklist
 | REQ ID | RTL Module | Status |
 |--------|-----------|--------|

 ## Findings
 ### [severity] Finding-N: ...

 ## Verdict
 PASS | FAIL: [reason]
 ```

7. Categorize issues: BLOCKER / WARN / SUGGESTION

• Any MISSING requirement is automatically a BLOCKER
• Any PARTIAL requirement is at minimum a WARN

<Tool_Usage>

Bash("mkdir -p reviews/phase-2-architecture")

Task(subagent_type="rtl-agent-team:rtl-architect",
     prompt="READ-ONLY review. (1) Read docs/phase-1-research/iron-requirements.json and check every REQ-NNN item for implementation in rtl/. Produce a Feature Coverage Checklist with per-REQ status: COVERED, PARTIAL, or MISSING. (2) Compare docs/phase-3-uarch/*.md against rtl/. List any spec-RTL mismatches, missing modules, or unspecified additions. Verify port naming follows project convention: i_ prefix for inputs, o_ prefix for outputs, io_ prefix for bidirectional. Save the review result to reviews/phase-2-architecture/architecture-review.md in standard review Markdown format with Date, Reviewer (rtl-architect, timing-advisor, rtl-critic), Upper Spec (docs/phase-1-research/iron-requirements.json, docs/phase-3-uarch/*.md), Verdict, Feature Coverage Checklist table, Findings, and Verdict sections. If the file already exists, update it with the latest review results. Output final verdict: VERDICT: PASS or VERDICT: FAIL — [N] spec violations found.")

Task(subagent_type="rtl-agent-team:timing-advisor",
     prompt="READ-ONLY review. Analyze rtl/ pipeline depth, clock domains ({domain}_clk naming, e.g. sys_clk), and reset strategy ({domain}_rst_n naming, e.g. sys_rst_n). Flag timing feasibility concerns and any clock/reset naming violations.")

Task(subagent_type="rtl-agent-team:rtl-critic",
     prompt="READ-ONLY review. Assess rtl/ code quality: synthesizability, coding style (lowRISC base with project overrides: i_/o_/io_ port prefixes, {domain}_clk/{domain}_rst_n naming, logic only, u_ instance prefix, gen_ generate prefix), maintainability. List issues with severity.")

</Tool_Usage>