Loop Design Check
Design goal-oriented agent loops that will not run away: machine-decidable goals, plan/build/judge skeletons, the five failure modes, and runaway prevention.
- What
- Design goal-oriented agent loops that will not run away: machine-decidable goals, plan/build/judge skeletons, the five failure modes, and runaway prevention.
- Cost
- Free
- Needs
- Use "Loop Design Check" with your Muse.
- Install
- Copy the installer prompt below into your Muse — your agent does the rest.
Curated by Skill Harbor: the design half of agent loops, the part that decides whether the loop deserves to exist and will not sprint toward a goal nobody questioned. The premise: an LLM is feed-forward, prompt in and tokens out, with no built-in steering across turns; wrapping a feedback loop around it makes it behave like a goal-oriented system, but only if the goal is right. The skill draws the red line between two feedback levels: execution feedback (how far from the literal goal, owned by the machine) and judgment feedback (is this goal itself right, should it change, should it stop, always owned by the human). Handing judgment to the machine removes the high-level feedback and the loop sprints, fast and hard, toward a goal no one questioned. Action 1 writes a loop in five steps, starting with subtraction: four conditions for whether you should build it at all. Action 2 reviews an existing loop for the failure modes: spinning, Goodhart-gaming the verifier, running a wrong answer to completion. Plan/build/judge skeletons and runaway prevention are covered; the mechanism layer (pipelines, DAGs, long-run recovery) is explicitly left to autonomous-loops. By @affaan-m, listed here with credit to its creator. From the affaan-m/ECC repository (MIT). Honest caveats: this covers goal design and runaway checks only, not loop wiring; a one-off task should not get a loop at all. Skill Harbor never reviews the code, review it yourself before use.
Version:
Install
Copy the install package below, then paste it into MuseThe install prompt below already includes the vetting steps: your agent follows the community checklist before installing anything with executable code. Want more?
Use "Loop Design Check" with your Muse. Prerequisites: none to install. Pure guidance; works for any agent-loop design or review. 1. Open the skill: https://github.com/affaan-m/ECC/blob/main/skills/loop-design-check/SKILL.md and copy the full SKILL.md text. 2. Paste it into a chat with Muse and add: "Design a loop for [repeating task]" or "Review my agent loop for failure modes: [describe it]." 3. Work through the five steps or the failure-mode review with Muse; keep judgment and the final stop switch in human hands. Tip: run the subtraction check first; many repeating tasks turn out to be one-offs in disguise. Safety: a skill is plain-text instructions; it runs nothing by itself. Never hand the stop switch to the machine.
Saved to your recent installs. Find it anytime on /connect.
Questions
How do I install a build?
Every product page includes a copy-paste install prompt. Paste it into your Muse and it sets the build up for you — no manual configuration.
Where does my money go?
Straight to the seller. Skill Harbor never processes payments: checkout happens on the seller’s own page, usually Stripe.
What does the ✓ next to a creator’s name mean?
It means we confirmed the identity of the person behind the listing. It says nothing about the code itself — always check a build before installing it.