Topic hub
Agent Harness
A New Runtime topic hub collecting signals, patterns, field notes, and public sources about agent harness.
Retrieval answer
A New Runtime topic hub collecting signals, patterns, field notes, and public sources about agent harness. Agent Harness is tracked here as an evidence-linked topic, not as a static glossary entry. The page connects raw observations to pattern hypotheses, longer analysis, and public sources. Use it as the canonical landing page before drilling into individual records.
Field notes
What should readers understand next?
A Software Factory Connects Agents Through Verified Outcomes
Augment and Warp describe team-level agent loops that move work from trigger and specification through implementation, verification, release, and measured improvement.
Agents Should Search, Fetch, And Browse As Separate Operations
Browserbase separates discovery, content retrieval, and browser interaction so research agents do not launch a full browser merely to obtain a list of URLs.
ChatGPT Cuts Repeated Work Across The Agent Stack
ByteByteGo's OpenAI engineering walkthrough connects persistent sessions, stable prompt prefixes, deferred tools, delta tokenization, cache-aware routing, and split inference.
Claude Code Auto Mode Gates Actions Instead Of Explanations
Claude Code Auto Mode combines an input injection probe with a two-stage action classifier, preserving autonomy while exposing an honest residual miss rate.
Cline Hooks Put Deterministic Rules Inside The Agent Loop
Cline's plugin hooks show how an agent harness can journal every run and block dangerous tool calls without waiting for the model to choose a guardrail.
Containment Caps An Agent's Blast Radius
Anthropic's three runtime patterns show why hard filesystem, network, credential, and trust boundaries carry more security weight than repeated approval prompts.
Anthropic Moves Large Tool Libraries Out Of Context
Anthropic's Tool Search Tool, Programmatic Tool Calling, and Tool Use Examples separate discovery, orchestration, and usage guidance for agents with large tool libraries.
Arcee Turns Scientific Post-Training Into A Run Ledger
Arcee's open-model science write-up shows a 21-run post-training loop around Trinity Mini, held-out scientific environments, trace review, and a promoted specialist adapter.
Genkit Adds Progressive Disclosure For Agent Skills
Genkit now loads Agent Skills through middleware that discovers SKILL.md metadata first and activates full instructions, references, and scripts only when needed.
Cline Turns Recursive Self-Improvement Into Harness Work
Cline's Terminal-Bench run is not a singularity story; it is a concrete loop where an agent reads traces, patches the harness, reruns evals, and hands a PR to humans.
OpenAI Shows Efficiency Is a Full-Stack Agent Problem
OpenAI's GPT-5.6 efficiency write-up connects model training, inference optimization, and the Codex/ChatGPT Work harness into one compounding cost-performance loop.
OpenAI's ARC-AGI-3 Jump Was a Harness Result
OpenAI's ARC-AGI-3 write-up shows why agent benchmarks measure the model plus the runtime harness: retained reasoning and compaction changed both score and token use.
Claude Opus 5 Makes Prompting a Runtime Contract
Anthropic's Opus 5 prompting guide shows that stronger models can make old harness defaults wrong: verbosity, effort, verification, delegation, and thinking mode all become runtime controls.
LangChain Deep Agents Shrink the Harness Instead of Adding More Prompt
Deep Agents v0.7.0b2 cuts default-agent input tokens by 65% and tool-description tokens by 43%, turning harness efficiency into a first-class agent metric.
Open Secure AI Alliance Turns the AI-Safety Fight Into a Stack Question
NVIDIA's Open Secure AI Alliance reframes open models, harnesses, identity, safe formats, scanners, and disclosure as shared defensive infrastructure for AI agents.
Agent Harness Optimization Is Becoming an Outer-Loop Discipline
The next useful automation target is not another agent response. It is the controlled loop that changes prompts, tools, context, and routing, then keeps only improvements that survive evaluation.
The Coding Harness Is Becoming Independent From the Model
Practitioners are routing different models through coding-agent workflows, while production systems increasingly choose model and effort per role instead of per product.
Raw signals
What changed recently?
Claude Opus 5 Turns Prompting Into Harness Calibration
Anthropic's Opus 5 prompting guide shows model upgrades changing the harness contract: verbosity, progress updates, scope, subagents, verification, and thinking mode all need retuning.
Anthropic: Goal-Scoped Agent Loops
The archive captures X source as a dated public record from X source. It documents long-running work gaining explicit goals, state, stopping rules, and recovery and is retained as pressure-testing evidence for the goal-scoped agent loops trend.
Claude Code: Verification Bandwidth
The archive captures Claude Code as a dated public record from Claude Code. It documents evaluation, review, and observability becoming the bottleneck after generation accelerates and is retained as pressure-testing evidence for the verification bandwidth trend.
GitHub / cursor/plugins: Verification Bandwidth
The archive captures GitHub / cursor/plugins as a dated public record from GitHub / cursor/plugins. It documents evaluation, review, and observability becoming the bottleneck after generation accelerates and is retained as pressure-testing evidence for the verification bandwidth trend.
Asteroid: Task-Specific Ai Interfaces
The archive captures Asteroid as a dated public record from Asteroid. It documents model output moving from plain answers into generated task interfaces and actions and is retained as supporting evidence for the task-specific AI interfaces trend.
Walkinglabs / Learn Harness Engineering: Verification Bandwidth
The archive captures Walkinglabs / Learn Harness Engineering as a dated public record from Walkinglabs / Learn Harness Engineering. It documents evaluation, review, and observability becoming the bottleneck after generation accelerates and is retained as pressure-testing evidence for the verification bandwidth trend.
Google / Gemini Models: Agent-Ready Software
The archive captures Google / Gemini Models as a dated public record from Google / Gemini Models. It documents software exposing explicit capabilities, permissions, and machine-readable actions and is retained as supporting evidence for the agent-ready software trend.
Claude: Goal-Scoped Agent Loops
The archive captures YouTube source as a dated public record from YouTube source. It documents long-running work gaining explicit goals, state, stopping rules, and recovery and is retained as supporting evidence for the goal-scoped agent loops trend.
SKILL.md Acts as a Behavior Loader, Not a Better Prompt
Skills package procedures, tools, and progressive context so an agent can load behavior only when the task requires it.
Claude: Goal-Scoped Agent Loops
The archive captures X source as a dated public record from X source. It documents long-running work gaining explicit goals, state, stopping rules, and recovery and is retained as pressure-testing evidence for the goal-scoped agent loops trend.
Stripe Built ProtoDash Around Its Own Design System
Stripe's internal prototyping harness combines product rules, components, sandboxes, and review feedback instead of generating generic mockups.
GitHub / NousResearch/hermes-agent: Generative Media Infrastructure
The archive captures GitHub / NousResearch/hermes-agent as a dated public record from GitHub / NousResearch/hermes-agent. It documents image, video, audio, and multimodal generation becoming application infrastructure and is retained as branch-opening evidence for the generative media infrastructure trend.
GitHub / stripe/link-cli: Task-Specific Ai Interfaces
The archive captures GitHub / stripe/link-cli as a dated public record from GitHub / stripe/link-cli. It documents model output moving from plain answers into generated task interfaces and actions and is retained as supporting evidence for the task-specific AI interfaces trend.
Ando: Ai-Native Operating Models
The archive captures Ando as a dated public record from Ando. It documents AI adoption shifting jobs, coordination, review, and organizational capacity and is retained as supporting evidence for the AI-native operating models trend.
Google Cloud / Gemini Enterprise Agent Platform: Ai-Native Operating Models
The archive captures Google Cloud / Gemini Enterprise Agent Platform as a dated public record from Google Cloud / Gemini Enterprise Agent Platform. It documents AI adoption shifting jobs, coordination, review, and organizational capacity and is retained as pressure-testing evidence for the AI-native operating models trend.
Image generation prompting Guide
The archive captures Image generation prompting Guide as a dated public record from OpenAI Developers / Image Gen Models Prompting. It documents image, video, audio, and multimodal generation becoming application infrastructure and is retained as branch-opening evidence for the generative media infrastructure trend.
Microsoft / New Hosted Agents In Foundry Agent Service: Verification Bandwidth
The archive captures Microsoft / New Hosted Agents In Foundry Agent Service as a dated public record from Microsoft / New Hosted Agents In Foundry Agent Service. It documents evaluation, review, and observability becoming the bottleneck after generation accelerates and is retained as pressure-testing evidence for the verification bandwidth trend.
Anthropic / 81k Economics: Verification Bandwidth
The archive captures Anthropic / 81k Economics as a dated public record from Anthropic / 81k Economics. It documents evaluation, review, and observability becoming the bottleneck after generation accelerates and is retained as pressure-testing evidence for the verification bandwidth trend.
GitHub / TencentCloud/CubeSandbox: Agent Security Runtime Boundaries
The archive captures GitHub / TencentCloud/CubeSandbox as a dated public record from GitHub / TencentCloud/CubeSandbox. It documents agent security expanding from prompt policy into memory, tools, sandboxes, and execution boundaries and is retained as branch-opening evidence for the agent security runtime boundaries trend.
Claude: Agent Security Runtime Boundaries
The archive captures YouTube source as a dated public record from YouTube source. It documents agent security expanding from prompt policy into memory, tools, sandboxes, and execution boundaries and is retained as branch-opening evidence for the agent security runtime boundaries trend.
Google / Ai Mode Chrome: Task-Specific Ai Interfaces
The archive captures Google / Ai Mode Chrome as a dated public record from Google / Ai Mode Chrome. It documents model output moving from plain answers into generated task interfaces and actions and is retained as supporting evidence for the task-specific AI interfaces trend.
GitHub / heygen-com/hyperframes: Generative Media Infrastructure
The archive captures GitHub / heygen-com/hyperframes as a dated public record from GitHub / heygen-com/hyperframes. It documents image, video, audio, and multimodal generation becoming application infrastructure and is retained as branch-opening evidence for the generative media infrastructure trend.
OpenAI / Next Evolution Of Agents Sdk: Agent-Ready Software
The archive captures OpenAI / Next Evolution Of Agents Sdk as a dated public record from OpenAI / Next Evolution Of Agents Sdk. It documents software exposing explicit capabilities, permissions, and machine-readable actions and is retained as pressure-testing evidence for the agent-ready software trend.
GitHub / farzaa/clicky: Task-Specific Ai Interfaces
The archive captures GitHub / farzaa/clicky as a dated public record from GitHub / farzaa/clicky. It documents model output moving from plain answers into generated task interfaces and actions and is retained as supporting evidence for the task-specific AI interfaces trend.