pipe

name	pipe
description	Chain multiple AI agent skills into sequential pipelines. Output of each skill feeds as context into the next.
trigger	/pipe
platforms	claude, codex, claw, opencode, aider, cursor, gemini

/pipe

Chain any skills together into a pipeline. Each skill's output becomes context for the next skill in the chain.

Usage

/pipe review -> security-review -> caveman-commit
/pipe analyze -> refactor -> test
/pipe explain -> document -> summarize
/pipe <skill1> -> <skill2> -> <skill3>

What /pipe Does

Without pipe, skills are isolated — /review has no idea /security-review exists. You manually copy context between them.

With pipe:

1. Runs skill1 fully
2. Captures output + context
3. Feeds into skill2 as prior context
4. Repeats down the chain
5. Final skill output = pipeline result

Rules

• Skills separated by -> (with spaces)
• Minimum 2 skills required
• Skills run sequentially, never in parallel
• Each step gets: original user request + all prior step outputs as context
• If a step fails or produces empty output, pipeline halts and reports which step failed
• No infinite loops — each skill runs exactly once per pipeline invocation

What You Must Do When Invoked

Parse the pipeline from the arguments. The format is:

/pipe <skill1> -> <skill2> -> ... -> <skillN>

Step 1 — Parse Extract skill names by splitting on ->. Trim whitespace. Minimum 2 skills. If fewer, tell user: "pipe needs at least 2 skills: /pipe skill1 -> skill2"

Step 2 — Announce Tell user the pipeline you're about to run:

Pipeline: skill1 → skill2 → skill3
Running step 1/3: skill1

Step 3 — Execute step 1 Look up what skill1 does from your knowledge of available skills or CLAUDE.md/AGENTS.md. Execute it fully as if the user invoked it directly. Capture the complete output.

Step 4 — Transition Announce: Step 1/3 complete. Running step 2/3: skill2 Inject prior step's output as context prefix: "Prior step (skill1) produced: [output summary]"

Step 5 — Execute step 2..N Repeat for each skill in chain. Each step receives:

• Original user request/codebase context
• Summary of all prior steps' outputs

Step 6 — Final summary After all steps complete:

Pipeline complete: skill1 → skill2 → skill3
[Final output from last skill]

Supported Platforms

/pipe works with any agent that supports skill/slash-command invocation:

• Claude Code (/pipe)
• Codex (via AGENTS.md)
• OpenCode (via AGENTS.md)
• Aider (via AGENTS.md)
• Claw (via AGENTS.md)
• Cursor (via .cursor/rules)
• Gemini CLI (via GEMINI.md)

Examples

Code review pipeline:

/pipe review -> security-review -> caveman-commit

Full review → security audit → compressed commit message, all in one command.

Understand before change:

/pipe explain -> refactor

Explains current code first, then refactors with that understanding as context.

Document everything:

/pipe review -> document -> summarize

Review code, generate docs, then summarize changes for PR description.