From autoloop
Manages autonomous coding loops: start runs with presets, monitor progress, inject guidance, merge/clean worktrees, inspect artifacts, and check health.
How this skill is triggered — by the user, by Claude, or both
Slash command
/autoloop:autoloop [run|status|guide|merge|clean|inspect] [args...][run|status|guide|merge|clean|inspect] [args...]The summary Claude sees in its skill listing — used to decide when to auto-load this skill
You are operating the **autoloop** CLI to manage autonomous coding loops. Parse `$ARGUMENTS` to determine the operation. If no arguments or ambiguous, ask what the user needs.
You are operating the autoloop CLI to manage autonomous coding loops. Parse $ARGUMENTS to determine the operation. If no arguments or ambiguous, ask what the user needs.
# Basic run with a preset
autoloop run <preset> "objective"
# Run in isolated worktree (recommended for code changes)
autoloop run <preset> "objective" --worktree
# Auto-merge worktree on completion
autoloop run <preset> "objective" --worktree --automerge
# Run a chain of presets in sequence
autoloop run --chain autocode,autoqa "objective" --worktree
Presets: autocode, autofix, autotest, autoqa, autoreview, autosimplify, autodoc, autosec, autoperf, autospec, autoresearch, autoideas, autopr, automerge
When to use --worktree: Always for code-modifying presets (autocode, autofix, autosimplify, autoperf). Optional for read-only presets (autoreview, autoideas, autosec).
When to use --automerge: When the user wants hands-off operation. Appends an automerge step that squash-merges on completion.
Backend override: Use -b claude, -b pi, -b kiro, -b claude-agent-acp, or -b acp:<provider>:<command> to switch backends.
Use run_in_background: true on the Bash tool when starting runs so the user isn't blocked.
# List active runs
autoloop loops
# List all runs (including completed/failed)
autoloop loops --all
# Show details for a specific run (supports prefix match)
autoloop loops show <run-id>
# Watch a run live (polls every 2s, exits on completion)
autoloop loops watch <run-id>
# Health summary (stuck/watching/failed runs)
autoloop loops health
autoloop loops health --verbose
When the user asks "how's it going" or "status" with no specifics, run autoloop loops first. If there's one active run, follow up with autoloop loops show <run-id>.
Send a message to a running loop. It appears once in the next iteration's prompt as ## OPERATOR GUIDANCE, then is consumed.
# Guide the latest active run
autoloop guide "Focus on error handling"
# Guide a specific run
autoloop guide --run <run-id> "Skip the auth module, focus on payments"
Use this when the user wants to steer a running loop without killing it.
# View the scratchpad (iteration-by-iteration summary)
autoloop inspect scratchpad
# View the prompt sent to backend for a specific iteration
autoloop inspect prompt <iteration> --format md
# View raw backend output for an iteration
autoloop inspect output <iteration>
# View coordination events
autoloop inspect coordination
# View topology (role routing graph)
autoloop inspect topology --format graph
# View metrics
autoloop inspect metrics --format terminal
# View memory
autoloop inspect memory
# All inspect commands support --run <run-id> for specific runs
Worktrees provide git-level isolation. Each --worktree run gets its own branch (autoloop/<run-id>).
autoloop worktree list
Shows: run ID, status, branch, base branch, merge strategy, created time. Orphaned worktrees (disk path missing) show (orphan).
# Merge with recorded strategy (default: squash)
autoloop worktree merge <run-id>
# Override merge strategy
autoloop worktree merge <run-id> --strategy squash # squash commits into one
autoloop worktree merge <run-id> --strategy merge # merge commit (no squash)
autoloop worktree merge <run-id> --strategy rebase # rebase onto base
If merge conflicts occur, the command aborts the merge, lists conflicting files, and exits non-zero. The user must resolve manually with git merge <branch> from the base branch.
# Clean orphaned + terminal (merged/failed/removed) worktrees
autoloop worktree clean
# Clean ALL worktrees including running ones
autoloop worktree clean --all
# Force-clean a specific run's worktree
autoloop worktree clean --force <run-id>
What gets cleaned:
git worktree remove (or force-delete the directory)git branch -d/-D to delete the branchWhat's skipped by default: Running worktrees are skipped unless --all or --force is used.
# Remove stale run directories older than 7 days
autoloop runs clean
# Custom age threshold
autoloop runs clean --max-age 30
autoloop run --worktree: branch autoloop/<run-id>, dir .autoloop/worktrees/<run-id>/tree/autoloop worktree merge <run-id> squashes into base branchautoloop worktree clean removes directory, branch, and metadataRoutine maintenance pattern:
# After reviewing and merging completed worktrees:
autoloop worktree clean
# Periodically clean old run state:
autoloop runs clean --max-age 14
# Start the web dashboard (default: http://127.0.0.1:4800)
autoloop dashboard
# Custom port/host
autoloop dashboard -p 3000 --host 0.0.0.0
# Show resolved config with provenance
autoloop config show
# Show as JSON
autoloop config show --json
# Show config file path
autoloop config path
# Memory operations
autoloop memory list
autoloop memory status
autoloop memory find "pattern"
autoloop memory add learning "insight"
autoloop memory remove <id>
# Task operations
autoloop task list
autoloop task add "description"
autoloop task complete <id>
| User wants to... | Command |
|---|---|
| Implement a feature | autoloop run autocode "..." --worktree |
| Fix a bug | autoloop run autofix "..." --worktree |
| Write tests | autoloop run autotest "..." --worktree |
| Review code | autoloop run autoreview "..." |
| Security audit | autoloop run autosec "..." |
| Write docs | autoloop run autodoc "..." --worktree |
| Check what's running | autoloop loops |
| Steer a running loop | autoloop guide "..." |
| See what a loop did | autoloop inspect scratchpad |
| Merge finished work | autoloop worktree merge <id> |
| Clean up | autoloop worktree clean && autoloop runs clean |
| Full hands-off | autoloop run autocode "..." --worktree --automerge |
npx claudepluginhub mikeyobrien/autoloop --plugin autoloopDiscovers, finds, audits, repairs, crafts, runs, debriefs, and publishes repeatable AI-agent loops. Analyzes code or coding threads for recurring work.
Designs and configures autonomous AI developer workflows (loops) with verification gates, persistent state, and stop conditions. Use when building agentic SDLC pipelines, self-iterating agents, or scheduled/automated development cycles.
Designs and runs scheduled discover-triage-implement-verify-escalate agent loops with risk-tier ladder, STATE/run-log/budget spine, and kill switch. Activates on outer loop, scheduled agent loop, PR/CI watch, dep-bump loop, or schedule trigger.