Skip to content
View cooleryu's full-sized avatar

Block or report cooleryu

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
cooleryu/README.md

cooleryu — AI Agent Reliability. Before the tool touches the world, make sure its evidence is still current.

ContextAdapt · Repositories · 中文

I work on making AI-agent failures observable, reproducible, and blockable.

Context safety · Tool calling · Observability · Evaluation · Long-running workflows

I like turning vague agent failures into small, reproducible tests—and then into guardrails that run before a tool changes the system.

01 / Building now

ContextAdapt blocks an action whose context and permission are no longer current.

An agent can keep a complete old plan after the user has replaced the task. The plan still looks valid, but acting on it can change the wrong environment or reuse permission that is no longer current.

ContextAdapt checks the action's provenance and current permission immediately before execution. Its core verifier runs locally and deterministically.

  • Try the synthetic failure case in 10 seconds—no install, API key, or model call required.
  • Gate execution with PASS, BLOCK, or fail-closed INSUFFICIENT_EVIDENCE results.
  • Block stale Codex and OMP tool calls before they modify a file.
  • Inspect the trace contract, cited evidence, and machine-readable integration result.

Repository · Action guard · Executor integrations · Machine-readable result

Alpha scope: ContextAdapt currently checks actions built from superseded context and protected edits without current permission. A PASS result is not a universal safety guarantee.

02 / Selected merged fixes

Project What changed
Microsoft Agent Framework #5893 Made Gemini honor a declarative outputSchema instead of treating it only as JSON mode.
PydanticAI #5443 Prevented AG-UI from stalling during multi-server MCP discovery.
IBM MCP Context Forge #4446 Avoided redirect-sensitive /mcp URLs in Streamable HTTP probes.
Chrome DevTools MCP #1960 Fixed timeouts when browser agents click a native <select> option.
OpenLIT #1139 Kept endpoint metadata on failed OpenAI chat spans.

03 / The thread

Agent behavior should be observable, reproducible, and grounded in current system state.

Popular repositories Loading

  1. My_resume My_resume Public

    Forked from billryan/resume

    TeX

  2. claude-code-l claude-code-l Public

    Forked from ultraworkers/claw-code

    Claude Code Snapshot for Research. All original source code is the property of Anthropic.

    TypeScript

  3. mcp-context-forge mcp-context-forge Public

    Forked from IBM/mcp-context-forge

    An AI Gateway, registry, and proxy that sits in front of any MCP, A2A, or REST/gRPC APIs, exposing a unified endpoint with centralized discovery, guardrails and management. Optimizes Agent & Tool c…

    Python

  4. chrome-devtools-mcp chrome-devtools-mcp Public

    Forked from ChromeDevTools/chrome-devtools-mcp

    Chrome DevTools for coding agents

    TypeScript

  5. openlit openlit Public

    Forked from openlit/openlit

    Open source platform for AI Engineering: OpenTelemetry-native LLM Observability, GPU Monitoring, Guardrails, Evaluations, Prompt Management, Vault, Playground. 🚀💻 Integrates with 50+ LLM Providers,…

    TypeScript

  6. cooleryu cooleryu Public

    GitHub profile README