Example Pipelines

The pipelines/examples/ directory includes several reusable patterns for standalone Ripline usage.

Included examples

• hello_world — minimal input → transform → output
• parallel_agents — multiple agent steps with explicit edges
• daily-brief — recurring summary-style workflow
• implement_story_claude / implement_story_codex — single coding task in a target repo
• spec_then_build_queue_claude / spec_then_build_queue_codex — spec first, then fan out implementation work
• write_tech_script_claude — longer-form content generation with file output

Claude Code vs Codex variants

• Claude Code variants use runner: claude-code and typically expect cwd or path inputs from the run input or a profile.
• Codex variants follow the same workflow shape with runner: codex.
• Both are intended to be copyable starting points, not framework-mandated patterns.

---

1. Implement Story

Purpose: Run a single development story: implement the described work in a codebase and commit.

• Input: One task with id, title, and detail.
• Claude/Codex: Use runner: claude-code or runner: codex with cwd: "{{ run.inputs.projectRoot }}". Set projectRoot in a profile or when invoking.

Usually invoked by the Spec → Build → Queue pipeline (spawn/enqueue), not run directly. To run alone for testing:

ripline run implement_story_claude --input '{"projectRoot":"/path/to/repo","task":{"id":"1","title":"Add login","detail":"Implement user login with email/password."}}'

Or use the dedicated Codex variant:

ripline run implement_story_codex --input '{"projectRoot":"/path/to/repo","task":{"id":"1","title":"Add login","detail":"Implement user login with email/password."}}'

---

2. Spec → Build → Queue

Purpose: Turn a high-level feature idea into a spec, decompose it into implementation stories, enqueue each story as a separate run of the Implement Story pipeline, then collect results and verify (build, tests, sense-check) once all are done.

• Input: task (string) — e.g. "Add a settings page with theme toggle and notification preferences."
• Flow: intake → spec (agent) → decompose (agent) → queue (agent, outputs JSON) → parse (transform) → spawn (enqueue) → collect (collect_children) → verify (agent) → result.
• Model split (optional): Use Sonnet for spec, decompose, queue, and verify (reasoning); use Haiku for the per-story implement step (speed). The doc examples in docs/pipelines/examples/ illustrate this.
• Spawn: The enqueue node creates one run of implement_story_claude or implement_story_codex per parsed task. The parent run pauses until all children are terminal (completed or errored). The collect_children node then aggregates child run results so the verify step can run build/tests and reason over outcomes.

All agent nodes use a built-in code runner. The example files ship with either runner: claude-code or runner: codex. Use a profile to set projectRoot so spawned runs and the verify step have the correct working directory.

Run:

# Standalone (Claude) — use a profile that sets projectRoot for spawned runs
ripline run spec_then_build_queue_claude --profile my-project --input '{"task":"Add a settings page with theme toggle"}'

Or use the dedicated Codex variant:

ripline run spec_then_build_queue_codex --profile my-project --input '{"task":"Add a settings page with theme toggle"}'

---

3. Write Tech Script

Purpose: Produce a full YouTube tech-script package from a video idea: technical brief → script draft → fact-check → save to disk.

• Input: idea (string, video concept/title), duration_minutes (number).
• Flow: intake → tech_brief (agent) → write (agent) → fact_check (agent) → save (agent, writes file) → result.
• Claude/Codex: Save node uses cwd: "{{ run.inputs.scriptsPath }}". Set scriptsPath in a profile or --input.

Run:

# Standalone (Claude) — profile supplies scriptsPath
ripline run write_tech_script_claude --profile my-scripts --input '{"idea":"How Ripline pipelines work","duration_minutes":12}'

---

Example profile for standalone code-runner variants

Use a profile to supply paths so you don’t repeat --input every time:

# ~/.ripline/profiles/my-project.yaml
name: my-project
description: "Main repo and scripts dir"
inputs:
  projectRoot: /path/to/your/repo
  scriptsPath: /path/to/your/scripts

Then:

ripline run spec_then_build_queue_claude --profile my-project --input '{"task":"Add dark mode"}'
ripline run write_tech_script_claude --profile my-project --input '{"idea":"Intro to Ripline","duration_minutes":8}'

These example pipelines live in pipelines/examples/ and are intended to be copy-paste-friendly starting points for standalone Ripline workflows.