Unverified50% confidenceFactExact time
The project still lacks the reader model (judging whether an arm can answer from what it retained) as well as the compression arm itself; this work is tracked in issue #41, and the README calls it the highest-leverage open task.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/16/2026
First Seen
Valid until: 10/30/2026
Sources
Related Entities
Related Claims
UnverifiedMost teams lack observability tools designed specifically for agents; traditional logging systems record code execution flows but cannot fully capture intermediate states such as prompts, reasoning chains, and tool call parameters in LLM invocations.71% similarUnverified真实项目的接口文档绝大部分缺乏响应内容示例，是AI测试落地的主要难点之一68% similarUnverifiedThe Harness has a "step completion tool" that requires the Agent to attach evidence when calling it (such as a document list); an empty list fails validation — using a programmatic threshold to turn intent into action and reducing the ambiguity brought by claiming completion in natural language alone.64% similarUnverifiedRequests involving tool definitions, structured output constraints, or image inputs have a hard capability floor: a model either handles them reliably or fails unpredictably, with no middle ground.64% similarUnverifiedAfter running multiple ticket-to-PR agents, a team concluded that the main reason coding agents perform poorly lies in the external framework (harness), not model capability.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1044523API
curl https://kongchang.com/api/v1/knowledge/claims/1044523MCP
get_claim(id=1044523)