gaasher/agent-loop-skills

---

Why loops-as-skills

Two ideas collided in late 2025, and this repo lives in the overlap:

• Skills became the portable unit. An Agent Skill is just Markdown + a little YAML that an agent loads only when relevant — “maybe a bigger deal than MCP … throw in some text and let the model figure it out” (Simon Willison). One SKILL.md now runs across ~30 hosts (Claude Code, Codex, Cursor, …).
• The loop became the program. Karpathy ran ~700 autoresearch experiments in 2 days from one markdown prompt; Geoffrey Huntley’s Ralph is, “in its purest form, a Bash loop.” Agents get most of their power not from one clever prompt but from iterating against feedback.

This repo makes the loop be the skill. Instead of task-specific skills, each entry is a generic loop — program · artifact · feedback signal · run ledger · termination — that you bind to your task at invocation time. Paste your goal; the loop proposes a change, runs it in your environment, scores it on a real signal (tests, latency, a metric, a calibrated judge), keeps it only if it’s better, logs it, and repeats.

The honest part: unsupervised agent loops are famous for spinning forever and confidently shipping garbage — at 90% per-step accuracy, a 5-step chain fails ~40% of the time. Every loop here is verification-gated: an objective feedback signal decides each step and an explicit termination condition ends it. That discipline — not autonomy for its own sake — is the point. (See Limitations.)

How a loop works

flowchart LR
  T["bind your task(artifact + signal + budget)"] --> P["proposeone change"]
  P --> R["run it inyour env"]
  R --> S{"scoretests · metric · judge"}
  S -->|better| K["keep + log"]
  S -->|worse| X["revert"]
  K --> G{stop?}
  X --> G
  G -->|"plateau · budget · threshold"| B(["best artifact"])
  G -->|no| P

Every loop decomposes into the same five ingredients — program (SKILL.md), artifact slot (what’s improved), feedback signal (what drives the next step), run ledger (append-only log), and termination (when to stop). Skills ship zero heavy dependencies: your code (a torch trainer, a SQL database, a dataset) runs in your environment via a bound run command; the skill shells out and reads the result. Multi-role loops use spawn-or-degrade — real isolated subagents on Claude Code, the same roles inline elsewhere.

Install

Any one of these installs all the loops:

Claude Code — plugin marketplace (add once, then install):

/plugin marketplace add gaasher/agent-loop-skills
/plugin install agent-loops@agent-loop-skills

Loops install namespaced as agent-loops: (e.g. agent-loops:karpathy).

Any Agent-Skills host — the standard installers:

npx skills add gaasher/agent-loop-skills                   # auto-detects host, installs to the right dir
gh skill install gaasher/agent-loop-skills --agent   # claude-code | codex | cursor | …  (--pin, gh skill update)

Manual — clone, then copy the loops into your host’s skills dir (pick the line for your host):

git clone https://github.com/gaasher/agent-loop-skills

cp -r agent-loop-skills/loops/* ~/.agents/skills/   # cross-tool: Codex, Cursor, Pi, OpenClaw, …
cp -r agent-loop-skills/loops/* ~/.claude/skills/   # Claude Code
# Hermes: hermes skills tap add gaasher/agent-loop-skills

Then just describe your task — the host loads the matching loop. Research loops also call the shared literature-search skill; installing everything puts it alongside them, and any loop degrades gracefully (to WebSearch) if it’s absent.

Loops in action

Most skill repos tell you what a skill is. Here’s what these loops actually do — real Sonnet runs, full ledgers in showcase/.

🧑‍⚖️ tournament-autoresearch — competing ideas, a self-calibrating judge

`` agents pitch competing changes each step; a judge critiques them, picks one, runs it, and recalibrates by comparing its predicted vs realized gain. On a CIFAR-10 SmallCNN under a fixed 5-epoch budget it climbed 0.734 → 0.798 val_acc, keeping 7 of 11 changes and reverting all 4 that regressed — escaping the plateaus a single-thread loop gets stuck on. (That’s the run charted up top.) → showcase/tournament-autoresearch

🔬 ml-autoresearch — analysis-first, every change traced to a cause

This loop reads inside each run — gradient flow, dead neurons, the loss curve — and grounds the next change in that evidence rather than guessing: “FC grad 57% vs first conv 3.3% — severe imbalance; 54% dead neurons” → add BatchNorm; “cosine schedule fixed the epoch-3 dip entirely (monotonic!), +0.033”. It also reverts what hurts (augmentation, over-aggressive LR). The point isn’t a leaderboard number — it’s that every accepted change has a measured reason behind it. → showcase/ml-autoresearch

📊 data-analysis — findings with a number behind every one

Hypothesis → verify, stdlib-only. On a planted dataset it surfaced 3 real findings and correctly refuted 2, with effect sizes matching ground truth and no hallucinations: enterprise vs consumer order value 184.90 vs 109.16 (Cohen’s d = 2.13), mobile return rate 32.8% vs 8.2% (RR 4.0) — and it reversed a plausible-but-wrong claim once it spotted a mobile confound. → showcase/data-analysis

The loops

† = multi-role (real subagents on Claude Code, inline elsewhere). Browse any folder for its SKILL.md.

Compatibility

Skills (a SKILL.md the model invokes) work broadly across the open standard. A loop dispatching a subagent from a role file at runtime is confirmed only on Claude Code — elsewhere multi-role loops (†) run their roles inline (still correct, just serial). Single-agent loops run fully everywhere.

Full, citation-backed matrix and the precise “why subagents are Claude-Code-only” reasoning: docs/compatibility.md. Off Claude Code, don’t rely on parallel subagent isolation.

Compose with other skill collections

These are self-contained, open-standard skills, so they coexist with any other collection — install both into the same skills dir and use them together. For example alongside K-Dense scientific-agent-skills:

npx skills add gaasher/agent-loop-skills              # these loops
npx skills add K-Dense-AI/scientific-agent-skills     # + a domain-skill library

They install as sibling folders (Claude Code namespaces each plugin; other hosts load all and pick by description). A loop’s analysis step can invoke any other installed skill — the same mechanism the research loops use to call literature-search.

Roadmap

Loops are most useful when they’re honest about what’s next. PRs on any of these are very welcome:

• [x] Blue-teaming + communication interface — shipped as blue-team (the defensive fixer) and purple-team (find → fix → re-verify), with the pull request as the communication interface to the target’s owner.
• [ ] Per-loop sandbox/ eval cases committed for every loop, so anyone can reproduce a run end-to-end.
• [ ] Host-specific subagent adapters (Cursor subagents, Hermes delegate_task) so multi-role loops get real isolation beyond Claude Code.
• [ ] More domains — eval-harness optimization, refactoring-at-scale, agent-trace debugging.

See open issues and grab a good first issue.

Status & limitations

Experimental — expect breaking changes. Pin a version if you need stability.

• Non-deterministic. Treat every output as a draft to verify, not a result to trust. Workaround: seed where supported; verify against your own oracle.
• No correctness guarantee. A loop can be confidently wrong. Workaround: every loop gates on a signal — keep a human in the loop and point it at checks you own.
• Host-dependent. Real subagent isolation is verified only on Claude Code. Workaround: see Compatibility.
• Cost & latency. Loops make many model calls. Workaround: start with a low iteration budget.

Non-goals (deliberate scope, not missing features): these are human-supervised loops, not fully-autonomous agents; not a model or runtime (bring your own host); not domain-exhaustive — for a very custom workflow, fork a loop, that’s what they’re for.

Contributing — and a note on open source 💜

I love open source, and this repo is built to be added to. You don’t need to be an expert and you don’t need to write code — a sharper description, a new loop, a bug report, or a pasted run transcript all make it better. New loops are welcome, and so are wild ideas.

Start with CONTRIBUTING.md and the authoring rubric in docs/skill-authoring-rules.md; grab a good first issue. Be kind, have fun, and if a loop helped you, a ⭐ genuinely helps others find it.

Provenance / credits

• karpathy — a faithful adaptation of Andrej Karpathy’s autoresearch program-as-skill idea.
• alpha-evolve — a simplified, ML-bent re-creation of AlphaEvolve (Novikov et al., 2025, arXiv:2506.13131) + OpenEvolve.
• research-proposal — its ScholarEval evaluator faithfully reimplements ScholarEval (arXiv:2510.16234).
• Layout and the dependency-isolation philosophy are informed by K-Dense scientific-agent-skills.

Repo layout

agent-loop-skills/
├── loops/            # one self-contained, installable skill per folder (SKILL.md + tools/roles/schemas/rubrics/examples)
├── showcase/         # real archived runs (ledgers + results) behind the examples above
├── assets/           # generated progress charts
├── docs/             # authoring rules, compatibility, api-keys, authoring quickstart
└── .claude-plugin/   # plugin.json + marketplace.json (Claude Code plugin / marketplace)