single-agent-loop-kit skill
Anthropic skill for Claude Code that audits a multi-agent system and, by measurement, folds subagents into one loop with skills.
- 506 ★ on the repository
- Last commit
- Apache-2.0
Extended access
Why this level
- The skill edits orchestration code in your repository and runs evals that spend paid model tokens.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/single-agent-loop-kitDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The skill edits orchestration code in your repository and runs evals that spend paid model tokens.
This is third-party code. Review the repository files before installing.
What it does
single-agent-loop-kit is an Anthropic kit with one skill, `single-agent-loop`, for code where a router or orchestrator hands work to subagents. The skill moves through steps: audit, baseline, merge, fold a subagent into a skill and tools, call a subagent as a tool, upkeep. Every change is scored on the same eval before and after. A subagent goes away only when the numbers say its replacement holds. The stops are fixed: after the audit, before any paid run, after the baseline and after each change. The repository also holds a teaching example, `examples/support-desk`: a support agent with a router and six subagents. It runs offline without an API key.
Who it is for. Teams maintaining a multi-agent system built on a router and subagents who want to check whether every subagent earns its place.
Good fit when
- A product has six subagents and it is unclear which to keep.
- Two subagents always run together and are candidates for merging.
- A subagent is turned into a skill and quality is compared before and after.
- No evals exist and a first set is built together with the agent.
Not a fit when
- You need a ready orchestration library rather than a stepwise review process.
- The system has a single agent and no subagents.
- Architecture will not be changed without a run on a real model.
Example request
Run the single-agent-loop skill: audit the subagents and show a verdict for each.Limitations
The repository is published as a reference implementation: its authors do not maintain it or accept contributions. Only two runs of the teaching example went through a real model; the merge and upkeep steps were not run that way. Paid runs spend tokens, so the skill stops before them with an estimate. Results depend on the model and the eval set, so conclusions for your system need your own measurements. The skill keeps its notes in `single-loop/` inside your repository, which you either commit or remove.
How to disable. Delete `.claude/skills/single-agent-loop` and the notes in `single-loop/` in your repository.
FAQ
Do I need an API key to try it?
No. The free check runs on a built-in stand-in model: `KIT_MODEL=fake python3 evals/run_eval.py --variant smoke` in `examples/support-desk`. It shows the plumbing works but gives no scores.
What does the skill do with subagents?
It gives each one a verdict: fold, merge, call as a tool or keep. You see the verdicts before any change.
Is the skill tied to a framework?
No. It reads whatever orchestration code is in the repository, whatever the framework.
What is the outcome?
Subagents that do not earn their own context become skills and tools of the main agent. The rest stay and are called as tools with a structured result.
Is the project maintained?
No. The README states that issues are not monitored; fork it and adapt it.
Discussion
No questions or reviews yet.
Related
Skills33 battle-tested skills and a minimal .claude harness for any agent. Installation by script, npx skills or cloning
AI Builder Club Skills
AI Builder Club — Skills
A Claude Code plugin marketplace to make a repo agent-ready and run compounding agent loops
A Claude Code skill and /looper command that interviews you and builds an agent loop with checks, an independent reviewer and stop rules before it runs
A Claude Code skill that takes a founder from idea to a live Claude Managed Agent: interview, v0 build, launch, grading and scheduling
A skill that turns several Claude Code, Codex and Cursor windows into a team with a coordinator, shared registry and cross-review
Loopy
Loopy (Loop Library)
A skill and library of ready agent loops: find, audit, repair, craft and run a repeatable workflow
README in short
The kit's principle is to measure before changing. A subagent stays when it earns its own context and folds only when scores hold.
README badge
Original SKILL.md
--- name: single-agent-loop description: Move a multi-agent system (router, planner, subagents) toward one agent loop with skills and tools, one measured change at a time. Use when asked to audit subagents, decide which to fold, merge, keep or call as tools, fold a subagent into a skill, or check whether a fold held up. ---