| name | analyzing-codebases |
| description | Detects project languages and monorepo state, runs language-appropriate static analysis (dependency graph, complexity, duplication, semantic patterns), and produces a refactor map ranking hotspots. Use when user invokes /aref or explicitly asks to analyze a codebase for refactoring. |
Analyzing Codebases
Overview
Analyzing codebases IS producing a hotspot-ranked refactor map before any restructuring begins.
Run the language-specific toolchain, aggregate results into a single map covering dependency graph, complexity hotspots, duplication, cyclic dependencies, and AGENTS.md gaps. The map is the input to planning-refactors; without it, planning is guesswork.
Core principle: Measure before touching code.
Routing
Pattern: Chain
Handoff: user-confirmation
Next: planning-refactors
Task Initialization (MANDATORY)
Before ANY action, create task list using TaskCreate:
- Subject:
[analyzing-codebases] Task N: <action>
Tasks:
- Detect languages and monorepo state
- Determine refactor scope (whole repo vs subproject)
- Check required toolchain availability
- Run toolchain and collect raw outputs
- Compute hotspot ranking (churn × complexity)
- Assemble refactor map
- Present map and await user approval
Task 1: Detect languages
Scan CWD for manifest files. Each manifest maps to a language:
package.json → TypeScript/JavaScript
pyproject.toml or setup.py → Python
Cargo.toml → Rust
go.mod → Go
Record detected languages in memory. No manifest → generic fallback.
Task 2: Determine scope
Check for monorepo markers:
pnpm-workspace.yaml, lerna.json, nx.json, turbo.json → JS monorepo
- Cargo
[workspace] in root Cargo.toml → Rust workspace
- Multiple
go.mod files → Go multi-module
- Multiple
pyproject.toml in subdirs → Python workspace
Monorepo → ask user: whole repo, specific subproject(s), or root only.
Single project → scope is CWD.
Task 3: Check toolchain
For each detected language, load the corresponding reference in references/. Check each tool with which <tool> or language-specific equivalent.
Missing tools → print install command from the reference; ask user to install-then-continue or skip that analysis class.
Task 4: Run toolchain
Execute the tools per reference instructions. Save raw outputs to .rcc/aref-raw/{ts}-{lang}-{tool}.<ext> where <ext> matches each tool's native output format (see Output Locations in each reference doc — .json, .csv, .txt, .dot etc.). {ts} = YYYYMMDD-HHMMSS, fixed for the run.
Task 5: Compute hotspots
Hotspot score = git log churn × cognitive complexity. Churn: git log --format=format: --name-only --since="6 months ago" | grep -v '^$' | sort | uniq -c | sort -rn. Complexity from toolchain output.
Rank top 20 files by score.
Task 6: Assemble refactor map
Write .rcc/{ts}-refactor-map.md per schema in references/refactor-map-schema.md.
Task 7: Present to user
Print summary: detected languages, scope, top 5 hotspots, count of cyclic deps, AGENTS.md status. Ask user: approve plan handoff / adjust scope / abort.
Approved → hand off to planning-refactors.
Red Flags - STOP
- "Skip toolchain, read code directly"
- "Compute hotspots from file size only"
- "Skip churn (no git history)"
- "Produce plan before map"
- "Scope = whole repo" on a monorepo without asking
Common Rationalizations
| Thought | Reality |
|---|
| "I can eyeball the hotspots" | Hotspots = churn × complexity. Both must be measured. |
| "Tool missing, skip silently" | Ask user. Silent skip produces incomplete map. |
| "Map too long, summarize aggressively" | Map is machine input for planning. Completeness beats brevity. |
| "Use LOC as complexity proxy" | LOC correlates weakly. Use cognitive complexity. |
| "Run on uncommitted changes" | Churn = git history. Dirty tree skews count. |
References
references/typescript-toolchain.md
references/python-toolchain.md
references/rust-toolchain.md
references/go-toolchain.md
references/generic-fallback.md
references/refactor-map-schema.md