Codemod Runner

Codemod Runner

Apply automated code transformations at scale using the right AST tool for the job — with scope assessment, batch execution, and idempotency verification before committing.

Tool Selection

flowchart TD
    Start([Codemod task received]) --> Q1{Language?}
    Q1 -->|Python| Q2{Transformation type?}
    Q1 -->|JavaScript / TypeScript| Q3{Needs full type info?}
    Q1 -->|Multi-language or<br>language-agnostic| Comby["comby<br>Structural search-replace<br>Works on any language<br>Pattern: comby 'old' 'new' .py"]
    Q2 -->|Rename, rewrite,<br>API migration| LibCST["LibCST<br>Preserves formatting<br>Python CST with type safety<br>pip install libcst"]
    Q2 -->|Lightweight search-replace<br>or regex-like patterns| Comby
    Q3 -->|Yes — needs type resolution,<br>cross-file inference| TSMorph["ts-morph<br>Full TypeScript compiler API<br>npx ts-morph-cli or use API<br>Best for typed refactors"]
    Q3 -->|No — structural only| Q4{Ecosystem preference?}
    Q4 -->|AST patterns preferred| AstGrep["ast-grep<br>Fast structural search<br>npm install -g @ast-grep/cli<br>sg run -p 'pattern' -r 'rewrite'"]
    Q4 -->|Codemod ecosystem,<br>existing transforms| JSCodeshift["jscodeshift<br>Facebook codemod runner<br>npx jscodeshift -t transform.js src/"]
    Comby --> Scope
    LibCST --> Scope
    TSMorph --> Scope
    AstGrep --> Scope
    JSCodeshift --> Scope
    Scope([Proceed to scope assessment])

Phase 1 — Scope Assessment (run before any transformation)

Assess impact size before touching files:

# Count affected files
rg -l 'old-pattern' | wc -l

# Preview first 25 affected files (saves to batch list)
rg -l 'old-pattern' | head -25 > batch.list
cat batch.list

# Count total occurrences (not just files)
rg -c 'old-pattern' | awk -F: '{sum+=$2} END {print sum}'

If the file count exceeds 50, process in batches using batch.list:

# Build per-batch lists of 25 files each
rg -l 'old-pattern' | split -l 25 - batch-
ls batch-*  # batch-aa, batch-ab, ...

Phase 2 — Batch Execution

Run the codemod against one batch at a time. Replace TOOL_CMD with the selected tool:

# ast-grep example
sg run -p 'old_fn($A)' -r 'new_fn($A)' --lang python $(cat batch-aa)

# comby example
comby 'old_fn(:[args])' 'new_fn(:[args])' -f .py $(cat batch-aa)

# jscodeshift example
npx jscodeshift -t transform.js $(cat batch-aa)

# LibCST — use a script, pass files as args
uv run codemod_script.py $(cat batch-aa)

# ts-morph — use a script targeting the batch
npx ts-node transform.ts --files $(cat batch-aa)

Phase 3 — Per-Batch Idempotency Check

After each batch completes, verify the transformation is idempotent (running it twice produces zero diff):

# Run the transformation a second time on the same batch
TOOL_CMD $(cat batch-aa)

# Confirm no further changes
git diff --stat
# Expected output: (empty — no changes)

If git diff shows changes after the second run, the transformation is not idempotent. Investigate the transform logic before proceeding to the next batch.

Phase 4 — Verification Trend

After each batch, measure progress toward zero remaining occurrences:

# Before starting:  rg 'old-pattern' | wc -l  → baseline count N
# After batch 1:    rg 'old-pattern' | wc -l  → should decrease
# After all batches: rg 'old-pattern' | wc -l  → 0

# One-liner to track trend across iterations
echo "Remaining: $(rg 'old-pattern' | wc -l)"

The count MUST decrease monotonically after each batch. If it does not decrease, the transform did not apply — investigate before continuing.

Final state: rg 'old-pattern' | wc -l returns 0.

Phase 5 — Commit

Once all batches pass idempotency and the verification count reaches 0:

git add -p   # or specific files
git commit -m "codemod: migrate old-pattern → new-pattern"

Commit after all batches complete and verification passes — not per-batch.

Tool Quick Reference

SOURCE: Tool official documentation — comby (comby.dev, accessed 2026-05-22), ast-grep (ast-grep.github.io, accessed 2026-05-22), jscodeshift (github.com/facebook/jscodeshift, accessed 2026-05-22), ts-morph (ts-morph.com, accessed 2026-05-22), LibCST (libcst.readthedocs.io, accessed 2026-05-22).

Large-Scale Codemod Swarm Pattern

For codebases with thousands of files, coordinate parallel workers across file batches using the /swarm-patterns skill (Pattern 6). Each worker owns one batch file, runs the codemod + idempotency check, and reports its completion count to the team lead before the final commit.