Glossary · term90 of 202
harness
The surrounding tooling and control logic that runs a model in a loop, deciding how it reads files, calls tools, and hands off work, separate from the model itself.
also written as agent harness · coding harness
Related terms
execution subagentA subagent given a narrow task with a fixed output shape and a written boundary, as opposed to open-ended reasoning or planning work. Its job is to follow the spec, not to expand scope.view termtasks primitiveA poll-based mechanism in MCP for durable work, letting a client fetch results after a dropped connection instead of resuming a live stream. It replaced resumable SSE streams as the way to survive a failure.view term
At a glance
- cited by
- 5 posts
- categories
- 3
- first used
- may 2026
Why expensive models make terrible execution subagentsSubagent model selection has one costly default. Reasoning models obey their own traces 15.6% of the time, so here is where to pin model and effort.
Every Model Release Ages Your SkillsClaude Code ignoring CLAUDE.md is often an aged file, not a model bug. See which instructions a release expires, and which lines to delete on day one.
Perplexity Isn't Shutting Down. Its Citation Corpus Might ShrinkNo, Perplexity is not down. The lawsuits threaten which sources it can read, so plan for a thinner citation set inside a healthy 200, and detect it in code.
How to Write Claude Code Skills That Actually TriggerWrite Claude Code skills that actually trigger: fix the description-as-router, structure for progressive disclosure, and test in fresh sessions.
Cut MCP Round-Trip Overhead by Looping Inside the ToolMCP round-trip overhead can balloon a task from 2,000 to 150,000 tokens. Move the loop inside one MCP tool to collapse N calls into a single round-trip.