Local-first workflows for agentic coding
Wire Codex, Claude, and shell commands into repeatable YAML pipelines. Run locally, version in Git, share with your team.
$
Works on macOS curl -fsSL https://tryrigg.com/install | bash 
Codex 
Claude Code 
opencode Soon 
Kimi Code Soon No more copy-pasting between agents
Wire your agents into a repeatable YAML pipeline. Define it once, run it with one command.
$
rigg run review-and-fix .rigg/review-and-fix.yaml
id: review-and-fix
steps:
- id: loop
type: loop
until: ${{ steps.judge.result.needs_fix == false }}
max: 5
steps:
- id: review
type: codex
with:
action: review
prompt: Review uncommitted changes for bugs.
output_schema:
properties:
findings: { type: array }
- id: judge
type: claude
with:
prompt: Evaluate findings and decide which need fixing.
output_schema:
properties:
needs_fix: { type: boolean }
- id: fix
if: ${{ steps.judge.result.needs_fix }}
type: codex
with:
action: exec
prompt: ${{ steps.judge.result.valid_findings }} 1
Codex reviews your changes Scans uncommitted changes and returns structured findings — bugs, style issues, potential regressions.
2
Claude judges the findings Evaluates each finding, filters out noise, and decides whether fixes are actually needed.
3
Codex auto-fixes, then loops Valid issues get fixed automatically. The loop repeats until clean — up to 5 iterations.
$
rigg run plan --input requirements="Add auth to the API" .rigg/plan.yaml
id: plan
inputs:
requirements: { type: string }
steps:
- id: draft
type: codex
with:
action: exec
prompt: Draft an implementation plan for: ${{ inputs.requirements }}
output_schema:
properties:
plan: { type: string }
- id: review
type: claude
with:
prompt: Review this plan for gaps: ${{ steps.draft.result.plan }}
output_schema:
properties:
has_issues: { type: boolean }
feedback: { type: string }
- id: refine
if: ${{ steps.review.result.has_issues }}
type: codex
with:
action: exec
prompt: Improve the plan based on: ${{ steps.review.result.feedback }}
- id: save
type: write_file
with:
path: PLAN.md
content: ${{ steps.refine.result.plan || steps.draft.result.plan }} 1
Codex drafts the plan Analyzes requirements and produces a structured implementation plan with file paths and steps.
2
Claude reviews for gaps A second AI reviews the plan for ambiguities, missing edge cases, and architectural issues.
3
Refine and save Refinement only runs when issues are found. The final plan is saved to PLAN.md.
$
rigg run parallel-checks .rigg/parallel-checks.yaml
id: parallel-checks
steps:
- id: checks
type: parallel
branches:
- id: security
steps:
- id: scan
type: codex
with:
action: review
prompt: Audit for security vulnerabilities.
- id: architecture
steps:
- id: review
type: claude
with:
prompt: Review architecture and design patterns.
- id: tests
steps:
- id: run
type: shell
with:
command: cargo test --workspace
- id: summary
type: claude
with:
prompt: Summarize all results: ${{ steps.checks.result }} 1
Three checks run at once Security audit, architecture review, and test suite run simultaneously — not sequentially.
2
Claude summarizes everything All parallel results flow into a single summary step. One command, full picture.
For Teams
Same workflow, every engineer
Drop .rigg/*.yaml into your repo. Every engineer runs the same
agent pipeline. Update workflows in PRs, review them like code.
No more "how do you use Claude for code review?" in Slack. The answer is in the repo.
project structure
your-project/
├── .rigg/
│ ├── plan.yaml # Claude plans → Codex builds
│ ├── review.yaml # Automated code review loop
│ └── fix-loop.yaml # Review → judge → fix cycle
├── src/
├── package.json
└── ...Start building agent workflows
Open source, written in Rust, runs locally. Three commands to your first workflow.
$
rigg init && rigg run plan --input requirements="..."