loop

Executes gated brainstorm-plan-build-review-improve cycles with durable resumable state and stop conditions.

Updated Sep 14, 2026
One-click install
npx skills add https://github.com/pandejesal/drone-nav-sar --skill loop-pandejesal
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: loop
Source: https://github.com/pandejesal/drone-nav-sar/tree/main/.swarm/bundled-skills/loop
Command: npx skills add https://github.com/pandejesal/drone-nav-sar --skill loop-pandejesal

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running a complex engineering objective end to end often stalls because work is unplanned, self-approved, or lost when a session is interrupted. This Skill runs an objective through gated, repeatable cycles with independent review and durable state so progress compounds instead of restarting. ## Core Features & Use Cases - Gated five-phase cycles: Runs BRAINSTORM, PLAN, BUILD, REVIEW, and IMPROVE phases, each with entry and exit gates backed by concrete evidence. - Generator/verifier separation: Delegates implementation, independent review, and critic challenge to separate contexts so no context approves its own work. - Durable resumable state: Persists cycle, phase, gate outcomes, and learnings under .swarm/loop/<run-id>/ so a killed session resumes without losing or redoing work. - Defense-in-depth stop conditions: Stops on objective met, cycle budget, plateau, oscillation, unrecoverable error, or explicit user request. - Use Case: Ask the agent to add rate limiting to a public API; it brainstorms a spec, plans tasks, implements them, passes independent review and critic gates, captures learnings, and loops until the success criteria pass or the budget is exhausted. ## Quick Start Run the loop mode on my objective to add rate limiting to the public API with at most three cycles and checkpoint autonomy.

Frequently Asked Questions about loop

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run an objective through the loop workflow?

Issue a loop command with a [MODE: LOOP ...] header containing your objective, plus optional max_cycles (1-5, default 3), autonomy (auto or checkpoint), depth, and resume flags. The skill then runs gated brainstorm, plan, build, review, and improve phases.

What is the difference between loop, full-auto, and turbo modes?

Loop is a user-initiated, gated, sequential workflow that plans, delegates, and executes through cycles. Full-auto is only a critic gate that reviews phase completions without planning or executing, while turbo runs parallel lanes within a single phase.

Can a loop run resume after the session is interrupted?

Yes. Loop state is persisted to .swarm/loop/<run-id>/state.json after every gate transition. Passing resume=true reloads the most recent unfinished run, validates its fields, and continues from the recorded cycle and phase.

When does the loop stop iterating?

The loop stops when the objective's success criteria pass, the cycle budget is exhausted, a cycle produces no qualifying improvement, a prior change is reverted (oscillation), an unrecoverable error occurs, or the user explicitly stops it.

Why does the loop require separate reviewer and critic contexts?

The generator/verifier separation rule prevents the coder context from approving its own diff. Independent reviewer and critic contexts examine the actual diff, and any edit after approval invalidates that approval and triggers re-review.