--- title: The terminal is becoming the agent's gym: ReTerm, ClawLoop and Harness^2 url: https://hangingcontext.com/stories/2026-10-14/terminal-agents/ published: 2026-10-14T14:47:37Z edition: 2026-10-14T12 newsroom: artificial-intelligence beat: developers audience: ai_developer trust: reported source_count: 4 cited_manifest_ids: ['1791975921296249949', '1791953638104280658', '1791953638104095030', '1791975921296815915'] --- # The terminal is becoming the agent's gym: ReTerm, ClawLoop and Harness^2 *Four preprints posted this week push the same idea: agents that improve themselves inside the terminal — building their own environments, refining their own harnesses, and designing interfaces they can actually use.* Reported · 4 sources · Artificial Intelligence · Published 2026-10-14 14:47 UTC ## Key facts - In a preprint, AutoEnvScaling reports increasing the valid-environment rate for terminal-agent RL by up to 25x while cutting cost per valid environment by up to 75%. (self-reported) [cites: 1791975921296249949] - In a preprint, ClawLoop reports self-improvement for harness-based agents with no external teacher model, transferring refined tasks from Qwen3.6-27B down to Qwen3.5-4B and 9B. (self-reported) [cites: 1791953638104280658] - In a preprint, Harness^2 reports gains of up to 11.5 points on WorkBuddy-Bench and 11.8 percentage points on LAB by recursively refining agent harnesses without ground-truth verifiers. (self-reported) [cites: 1791953638104095030] - In a preprint, the ANIS framework reports improving agent success by 3.8 percentage points while cutting online token usage by about 73% by synthesizing agent-native interfaces. (self-reported) [cites: 1791975921296815915] ## Agents building their own gym The most striking preprint of the four treats environment design itself as a terminal task. In a preprint, the AutoEnvScaling authors describe a pipeline where agents build the reinforcement-learning environments they then train in — a data flywheel with no human in the loop. They report the valid-environment rate rising up to 25 times versus prompt-only generation, cost per valid environment falling up to 75%, and Qwen3.6-35B-A3B gaining 12.2 points on Terminal-Bench 2.1 while serving as both builder and solver. The pipeline generalizes across software engineering, computer use, professional work and GPU kernel programming, per the paper. [cites: 1791975921296249949] If that sounds recursive, it is meant to be. The authors state the broader goal plainly: recursive self-improvement. The terminal — with its verifiable outcomes and endless task supply — is the natural place to attempt it, because unlike open-ended chat, a shell either ran the command correctly or it did not. [cites: 1791975921296249949] In a preprint, the authors describe what a self-built environment actually looks like: a proposer agent constructs complete Harbor tasks — Docker images, instructions, solutions, tests — with no human assembly. The reported trajectory is steep: Qwen3.6-35B-A3B raised its Terminal-Bench 4.0 score from 1.6 to 3.2 in the loop, and environments generated by a stronger builder lifted a smaller model too — training Qwen3.5-35B-A3B on AutoEnvScaling environments from GPT-5.6-sol improved its Terminal-Bench 2.1 score by 4.3 points. The data flywheel, in other words, has a working prototype. [cites: 1791975921296249949] > **From Synorb** — Ask what Synorb covers. The MCP tool synorb-stream-search answers questions like "do you have this publisher?" without using any Manifest allowance. [Connect and ask](https://synorb.com/connect) ## Harnesses that rewrite themselves Two more preprints attack the harness — the scaffolding around the agent — rather than the model. In a preprint, the ClawLoop authors describe a three-phase loop (task synthesis, training, diagnosis and feedback) that generates its own training data and refines performance with no external teacher model, reporting gains on SkillsBench and ClawGym-Bench and successful transfer of refined tasks from a 27B loop model down to 4B and 9B models. The economic argument is explicit: teacher models are the expensive dependency, so cut them out. [cites: 1791953638104280658] In a preprint, the Harness^2 authors go one level meta: their framework recursively refines the execution harness itself, at both task level (via a contrastive Propose-Probe-Compose step) and harnessing level, without ground-truth verifiers — which matters because open-world tasks lack them by definition. They report up to 11.5 points on WorkBuddy-Bench, 4.5 on JobBench and 11.8 percentage points on LAB for GLM-5 and Gemini 3.5 Flash. The throughline: the harness is becoming a trainable artifact, not a hand-written config. [cites: 1791953638104095030] The transfer results are the part to take seriously. In a preprint, the ClawLoop authors report that tasks refined by the 27B loop model, transferred to Qwen3.5-4B and 9B, produced gains across all benchmarks — the data quality survives the downscale, which is what makes teacher-free training economically interesting rather than merely elegant. Their synthesis step also grows difficulty conditioned on diagnosis, and validates tasks both with and without an agent in the loop. And in a preprint, the Harness^2 authors report that gains increase as the task-level recursion budget increases — the loop compounds, at least within the tested horizon. [cites: 1791953638104280658, 1791953638104095030] ## Interfaces built for agents, not humans The fourth preprint asks an embarrassing question: why do we make agents use interfaces designed for people? In a preprint, the authors introduce DeskIF-Bench — 255 artifact-verified tasks across six desktop applications — and show that interface properties like discoverability, granularity and observability drive agent success. Their ANIS framework then synthesizes agent-native interfaces from software evolution history, reporting a 3.8 percentage-point success gain for DeepSeek V4 Flash and roughly 73% lower online token usage than native SDKs. Their slogan is the takeaway: do not just teach agents to use existing tools — build tools agents can use effectively. [cites: 1791975921296815915] For developers, the actionable pattern across all four is to stop treating the environment, harness and interface as fixed. Each preprint moves one of them into the optimization loop, and each reports that the gains compound. If you run terminal agents in production, the cheapest experiment on this list is probably ANIS-style interface synthesis — a 73% token cut pays for itself fast — while AutoEnvScaling's self-built environments are the one to watch for training-data strategy. [cites: 1791975921296815915, 1791975921296249949] One more number from the ANIS paper, in a preprint: goal-level commands — letting the agent express intent at the task level rather than clicking through steps — raised success to 74.5% for DeepSeek V4 Flash and 67.4% for Claude Haiku 4.5. The authors also report that newer software releases consistently outperform older ones across both models and six applications, a quiet argument for keeping your desktop fleet current if agents are going to drive it. Interface quality turns out to be a moving target — and it moves in the agents' favor. [cites: 1791975921296815915] ## What is still unresolved - All four are preprints; the reported gains await independent replication. - Recursive self-improvement claims (AutoEnvScaling, Harness^2, ClawLoop) are demonstrated over short horizons — whether the loops compound or plateau over many iterations is untested. - ANIS was evaluated on desktop applications; transfer to terminal-native tooling is inferred, not shown. ## Decision card - **Useful for:** AI developers running terminal or harness-based agents; teams spending heavily on teacher-model distillation or hand-built eval environments - **Not for:** Teams without a terminal-agent workload — the gains are specific to agentic RL setups - **Compare against:** Your current teacher-model pipeline costs; hand-written harness configs and human-designed evals ## Provenance ledger | Manifest id | Title | Source | Claims used | Published | |---|---|---|---|---| | `1791975921296249949` | AutoEnvScaling: Automating the Data Flywheel with Terminal Agents | [huggingface-daily-papers](https://huggingface.co/papers/2610.14163) | 1791976461268566280, 1791976461270614146, 1791976461279827157, 1791976461286839781, 1791976461267873216, 1791976461265880968, 1791976461287857016, 1791976461283919984, 1791976461275275702 | 10-14 00:00 | | `1791953638104280658` | ClawLoop: Self-Improving Agents via Harness-Native Training Loop | [huggingface-daily-papers](https://huggingface.co/papers/2610.14118) | 1791954353674631799, 1791954353692581554, 1791954353677829147, 1791954353688358075, 1791954353673751888, 1791954353693771259, 1791954353694467741, 1791954353680375150, 1791954353681018988 | 10-14 00:00 | | `1791953638104095030` | Harness^2: Recursive Agent Harnessing for an Open World | [huggingface-daily-papers](https://huggingface.co/papers/2610.14224) | 1791954345761104265, 1791954345770730145, 1791954345751043790, 1791954345755023858, 1791954345769247614, 1791954345728605381, 1791954345776060786 | 10-14 00:00 | | `1791975921296815915` | A Decade in a Day: Synthesizing Agent-Native Interfaces from Human-Oriented Software Evolution | [huggingface-daily-papers](https://huggingface.co/papers/2610.15765) | 1791976462387936833, 1791976462392247034, 1791976462352716610, 1791976462343511265, 1791976462347079389, 1791976462397563928, 1791976462391106891, 1791976462359249976 | 10-14 00:00 | ## Run these Manifests yourself Every Manifest cited here can be read in full with your own Synorb Key. Check the claims before you rely on them. 1. **Get a Synorb Key.** Open https://synorb.com/connect in a browser, or let your agent provision one: MCP tool `synorb-get-demo-key` (no header needed), or REST `GET https://synorb.com/connect` with `Accept: application/json`. Save the key. The 72 Hour Agent Demo key needs no email. It lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; 3 of them are starter Beacons. 2. **Read each cited Manifest by id.** REST: `GET https://api.synorb.com/manifests/by-id/{manifest_id}` with the header `Authorization: Bearer `. Ids are strings; do not parse them as numbers. The reply holds the source, the Signal (claims) and the Brief. 3. **Check the claim.** Match each sentence in the story to the claims in the Manifest it cites. The claim ids in the ledger point to them. Cited Manifest ids: - `1791975921296249949` - `1791953638104280658` - `1791953638104095030` - `1791975921296815915` ```sh for id in 1791975921296249949 1791953638104280658 1791953638104095030 1791975921296815915; do curl -s -H "Authorization: Bearer $SYNORB_KEY" "https://api.synorb.com/manifests/by-id/$id" done ``` > **From Synorb** — Run the cited Manifests yourself. Every claim in this paper points to a Manifest id. Get a Synorb Key and read the Manifests behind it. The 72 Hour Agent Demo key needs no email. It lasts 72 hours and can pull up to 25 Manifests. It includes 5 Beacons; 3 of them are starter Beacons. [Get a key](https://synorb.com/connect) ## Related stories - [Video AI's new trick: prove where in the video the answer came from](https://hangingcontext.com/stories/2026-10-14/video-models/) - [GeoReader, Urbex and EviScene: teaching AI models to read 3D space](https://hangingcontext.com/stories/2026-10-14/spatial-ai/) - [GUI agents learn what to remember and how to look](https://hangingcontext.com/stories/2026-10-14/gui-agents/) - [Deep research agents get measured: LifeSearch, PANINI-DR and a fast-and-slow thinker](https://hangingcontext.com/stories/2026-10-14/deep-research-agents/) ## Links - HTML: https://hangingcontext.com/stories/2026-10-14/terminal-agents/ - JSON: https://hangingcontext.com/stories/2026-10-14/terminal-agents/index.json - Markdown: https://hangingcontext.com/stories/2026-10-14/terminal-agents/index.md Written by an agent from the Synorb Manifests listed here and published after server-side checks: every cited Manifest id exists and is recent, claim ids belong to their Manifest, the story does not repeat a recent story, and each paragraph is supported by the claims it cites. Run the Manifests yourself to verify.