Chain

Chain Runner

You are executing a workflow chain. Chains are multi-step sequences defined in .claude/chains.json that run skills, read files, commit code, or update memory — in order, with optional pause points for human review between steps.

How It Works

A chain is a named sequence of steps. Each step is one of:

• Invoke a skill — run another Claude Code skill by name
• Read files — read specified files and summarize them
• Commit — stage and commit changes with a session summary
• Update memory — write results from previous steps to memory
• Summarize — produce a summary of findings so far

Steps can pause for review, letting you approve, skip, or stop.

Step 1: Load the Chain

Read .claude/chains.json. If the user specified a chain name, load it. If not, list available chains with descriptions and ask which to run.

If .claude/chains.json doesn't exist, tell the user: "No chains defined yet. Use /chain-create to build one, or create .claude/chains.json manually."

Step 2: Execute Steps in Order

For each step in the chain:

1. Announce: "Step N/Total: [description]"
2. Execute based on step type:
   • "skill" → invoke that skill by name
   • "action": "read" → read specified files, summarize contents
   • "action": "commit" → git add + commit with session summary
   • "action": "update-memory" → update memory file with results from previous steps
   • "action": "summarize" → summarize findings from previous steps
3. If "pause": true → show results, ask: "Continue? (y/skip/stop)"
   • y — continue to next step
   • skip — skip next step, move to the one after
   • stop — end chain here

Step 3: Summary

After all steps complete (or chain is stopped):

Chain: [name] — [completed / stopped at step N]
Steps run: N/Total
Issues found: [count from audit steps, if any]

Step 4: Suggest Next

Check suggest_next in chains.json. If the last skill run has suggestions:

Suggested next: /[skill-name] — [reason]

Rules

• Never skip steps silently
• Always show step progress (Step 2/4, etc.)
• If a skill fails, report it and ask whether to continue or stop
• Don't modify chains.json — use /chain-create for that

Example: chains.json

{
  "chains": {
    "post-code": {
      "description": "After code changes — review, audit, commit",
      "steps": [
        { "skill": "simplify", "description": "Review code quality", "pause": true },
        { "skill": "launch-readiness", "description": "Check infrastructure gaps", "pause": true },
        { "action": "commit", "description": "Commit if all clean" }
      ]
    },
    "start-session": {
      "description": "Get context at start of work",
      "steps": [
        { "action": "read", "files": ["README.md", "CLAUDE.md"], "description": "Read project context" },
        { "action": "summarize", "description": "What needs doing today" }
      ]
    }
  },
  "suggest_next": {
    "simplify": ["launch-readiness"],
    "launch-readiness": ["save"]
  }
}