Rust Crate Review

Rust Crate Review

Architectural health check for a single Rust crate and its one-hop neighborhood (direct dependencies + consumers). Produces a saved review document and a set of open questions for systems-thinking dialogue.

Not a PR review — use requesting-code-review for diff-scoped review. This skill reasons about design intent and responsibility boundaries over time, not diffs.

When to Use

• After a period of heavy AI-authored changes to a crate
• Periodic maintenance check (monthly or per-release cycle)
• Onboarding a junior contributor who needs to build a mental model of a crate

Invocation

/rust-crate-review <crate-name>

<crate-name> is the directory name under crates/ (e.g., bloqade-lanes-dsl-core, bloqade-lanes-search).

Step 1 — Automated Harvest (no agents)

Run these four commands and collect all output before spawning any agents. This is fast (seconds) and gives every agent a reliable structural foundation.

Before running command 3, derive <crate-module> from <crate-name> by replacing dashes with underscores (e.g., bloqade-lanes-search → bloqade_lanes_search).

# 1. Workspace dependency graph
cargo metadata --format-version 1 --no-deps

# 2. Change hotspots — 30-day window scoped to the crate directory
git log --since="30 days ago" --name-only --pretty=format:"%h %an %s" \
  -- crates/<crate-name>/

# 3. Consumer file paths across the workspace
#    Matches `use <crate-module>`, `<crate-module>::Foo`, and
#    `use <crate-module> as bar` forms; excludes the crate's own sources.
rg "\b<crate-module>\b" --type rust -l \
  | grep -v "^crates/<crate-name>/"

# 4. AI-authored commits in the harvest window
#    `Co-Authored-By:` lives in commit trailers, so this needs its own pass.
git log --since="30 days ago" --grep="Co-Authored-By: Claude" \
  --name-only --pretty=format:"%h %an %s" -- crates/<crate-name>/

Default to --since="30 days ago" for commands 2 and 4. After running command 2, check the count and rerun both if needed:

• Output empty or under 5 commits → rerun with --since="90 days ago"
• Output exceeds 30 commits → rerun with --since="14 days ago" to focus on recent drift

From the harvest output, extract before proceeding:

• Hotspot files: from command 2, files appearing in 3+ commits → pass to Agent 3 for weighted scrutiny
• AI-authored commits: command 4 already filters to these — pass the commit list (with file names) to Agent 3 for the AI-drift lens
• Large non-AI commits (secondary signal): from command 2, any commit touching 10+ files whose subject matches ^feat|^refactor|^chore → also flag to Agent 3

Step 2 — Phase 1: Parallel Agents

Dispatch Agent 1 (External API Mapper) and Agent 2 (Internal Architecture Mapper) simultaneously using the dispatching-parallel-agents skill.

Pass to both agents:

• The target crate name
• The full harvest output from Step 1

Use the prompt templates in agent-prompts.md — fill in {{CRATE_NAME}} and {{HARVEST_OUTPUT}} before dispatching.

Step 3 — Phase 2: Sequential Agent

Once both Phase 1 agents have returned, dispatch Agent 3 (Critical Evaluator).

Pass to Agent 3:

• The target crate name
• The full harvest output from Step 1
• The complete output from Agent 1
• The complete output from Agent 2

Use the Agent 3 prompt template in agent-prompts.md — fill in all four placeholders: {{CRATE_NAME}}, {{HARVEST_OUTPUT}}, {{AGENT_1_OUTPUT}}, {{AGENT_2_OUTPUT}}.

Step 4 — Synthesis

Assemble the final review document from the three agent outputs following this structure:

# Crate Review: <crate-name> (<YYYY-MM-DD>)

## 1. Context
   One-paragraph summary of what the crate does, its dependency position
   (what it depends on, what depends on it), and change activity level.

## 2. External API Surface
   From Agent 1: public type inventory, responsibility portraits,
   API friction points, dead public surface.

## 3. Internal Architecture
   From Agent 2: module map, internal interaction graph,
   pub(crate) type inventory, coupling hotspots, responsibility portraits.

## 4. Critical Evaluation
   From Agent 3: contract divergence, Rust health findings (hotspot-weighted),
   architectural health, AI-drift findings, ⚠ emerging pattern callouts.

## 5. Open Questions
   One subsection per finding area that produced findings.
   ### Contract Divergence
   ### Rust Health
   ### Architectural Health
   ### AI-Drift
   ### Emerging Patterns

Save to: docs/superpowers/reviews/YYYY-MM-DD-<crate-name>-review.md (create the docs/superpowers/reviews/ directory if it does not yet exist — on first run, only plans/ and specs/ are siblings).

After saving, surface the Section 5 questions directly in the conversation as a prompt to the user — do not just leave them buried in the document.

Emerging Pattern Callout Format

Agent 3 emits these as structured blocks. Preserve them verbatim in Section 4:

⚠ Emerging Pattern: "<pattern name>"
  Appears in: <file:line>, <file:line>, <file:line>
  Similarity: <description of structural resemblance>
  Signal: <N> instances, last added <X> days ago
  Suggested abstraction: <trait name / fn signature>
  Readiness: [ready to abstract | still evolving | monitor]

Readiness thresholds (from git log recency):

• All instances stable 20+ days → ready to abstract
• Any instance touched within 7 days → still evolving
• Between 7–20 days → monitor

Prompt Templates

See agent-prompts.md for the complete Agent 1, Agent 2, and Agent 3 prompt templates with placeholder conventions.

Source: QuEraComputing/bloqade-lanes — distributed by TomeVault.