pipeline-runner

description	Execute multi-step pipelines defined in JSON with:
name	pipeline-runner
triggers	["run pipeline","create pipeline","workflow","multi-step","backtest and deploy","if this then that"]
category	development
maturity	stable
tags	["pipeline","workflow-orchestration","dependency-graph","retries","json"]

Pipeline Runner Skill

Overview

Execute multi-step pipelines defined in JSON with:

• Sequential and parallel stages
• Dependency chains (needs)
• Conditional execution (if)
• Retries with backoff
• Timeouts per step
• Variable passing between steps (stdout → env var)
• Failure handling (on_fail: continue|notify)
• Run history and logs

Script

node ${CLAUDE_SKILLS_DIR:-$HOME/.claude-agent/.claude/skills}/pipeline-runner/scripts/pipeline.cjs <command> [args]

Commands

pipeline run <file.json> [--dry-run] [--var key=value]   — Execute a pipeline
pipeline validate <file.json>                             — Validate syntax
pipeline list                                             — List saved pipelines
pipeline history [--limit 10]                             — Recent runs
pipeline show-run <run-id>                                — Run details

Pipeline Format (JSON)

{
  "name": "my-pipeline",
  "env": { "MY_VAR": "value" },
  "steps": [
    {
      "name": "build",
      "run": "npm run build",
      "timeout": 120,
      "retries": 2,
      "retry_delay": 5
    },
    {
      "name": "test",
      "needs": ["build"],
      "run": "npm test",
      "on_fail": "continue"
    },
    {
      "name": "deploy",
      "needs": ["test"],
      "if": "steps.test.exitCode == 0",
      "run": "./deploy.sh"
    },
    {
      "name": "checks",
      "parallel": [
        { "name": "lint", "run": "npm run lint" },
        { "name": "typecheck", "run": "tsc --noEmit" }
      ]
    }
  ]
}

Step Properties

Property	Type	Default	Description
name	string	required	Unique step identifier
run	string	—	Shell command to execute
needs	string[]	—	Steps that must complete first
if	string	—	Condition: steps.NAME.exitCode == 0
timeout	number	300	Seconds before kill
retries	number	0	Retry count on failure
retry_delay	number	5	Seconds between retries
on_fail	string	halt	continue or notify or halt
parallel	step[]	—	Run sub-steps in parallel

Variable Passing

Each step's stdout is available to subsequent steps as: $STEP_<NAME>_OUTPUT (name uppercased, non-alphanumeric → underscore)

Agent Workflow

Creating pipelines on the fly

When the user asks for a multi-step workflow, write a JSON pipeline file and run it:

# Write pipeline
cat > /tmp/my-pipeline.json << 'EOF'
{ "name": "...", "steps": [...] }
EOF

# Execute
node pipeline.cjs run /tmp/my-pipeline.json

Saving reusable pipelines

Store in ~/.pipelines/ for future use:

cp /tmp/my-pipeline.json ~/.pipelines/backtest-flow.json

Dry runs

Always dry-run complex pipelines first:

node pipeline.cjs run pipeline.json --dry-run

Examples

See ${CLAUDE_SKILLS_DIR:-$HOME/.claude-agent/.claude/skills}/pipeline-runner/examples/:

• test-pipeline.json — basic test with all features
• backtest-pipeline.json — prop-hedge backtest → analyze → commit → push

Storage

• Pipeline files: ~/.pipelines/ (JSON)
• Run logs: ~/.pipelines/runs/ (JSON, one per run)