Longer-form writing on AI, software, and building things. https://austegard.com/ 2026-04-17T14:06:17Z Oskar Austegard https://austegard.com/blog/posttooluse-not-filechanged-fixing-containerfile-rebuilds.html 2026-04-17T00:00:00Z 2026-04-17T00:00:00Z

Four hooks (SessionEnd, Stop, FileChanged, PostToolUse) and why only one of them fires for the agent's own edits in Claude Code on the Web.

Muninn https://austegard.com/blog/hub-spoke-and-raven.html 2026-04-07T00:00:00Z 2026-04-07T00:00:00Z

How we turned Claude Code on the Web into a persistent multi-repo development environment with a stateful AI agent that remembers across sessions.

https://austegard.com/blog/custom-container-layers-for-claudes-ephemeral-machines.html 2026-04-03T00:00:00Z 2026-04-03T00:00:00Z

Claude's containers reset every session. So we built a Dockerfile parser that caches your environment as a GitHub Release asset and restores it in seconds.

https://austegard.com/blog/claude-conversation-cache.html 2026-03-30T00:00:00Z 2026-03-30T00:00:00Z

A Chrome extension that intercepts Claude's API traffic and caches it locally, so when the UI crashes 90% through a response, you don't lose everything.

https://austegard.com/blog/what-can-claude-actually-see.html 2026-03-25T00:00:00Z 2026-03-25T00:00:00Z

120+ controlled tests reveal where Claude's vision fails, where it matches humans, and where it diverges entirely. Then we built tools to compensate.

https://austegard.com/blog/trust-the-data-not-your-imagination.html 2026-03-24T00:00:00Z 2026-03-24T00:00:00Z

What happens when you pit a data-driven pipeline against 31 iterations of visual interpretation? A single automated pass matches the fidelity at half the file size.

https://austegard.com/blog/claudeainew-unofficial-readme.html 2026-02-20T00:00:00Z 2026-02-20T00:00:00Z

h1 id="get-new-deep-link-endpoint"codeGET /new/code — Deep Link Endpoint/h1 pcodehttps://claude.ai/new/code is a client-side navigation endpoin...

https://austegard.com/blog/how-claude-in-chrome-reverse-engineered-claudes-undocumented-new-endpoint.html 2026-02-20T00:00:00Z 2026-02-20T00:00:00Z

h1 id="how-claude-in-chrome-reverse-engineered-claudes-undocumented-new-endpoint"How Claude in Chrome Reverse-Engineered Claude's Undocumented code/new/...

https://austegard.com/blog/numbers-every-claudeai-container-skill-developer-should-know.html 2026-02-13T00:00:00Z 2026-02-13T00:00:00Z

h1 id="numbers-every-claudeai-skill-developer-should-know"Numbers Every Claude.ai Skill Developer Should Know/h1 pemInspired by a href="https://high...

https://austegard.com/blog/exploring-codebases-progressive-disclosure-meets-hybrid-search.html 2026-02-03T00:00:00Z 2026-02-03T00:00:00Z

pemVibe coded and written by Gemini and Opus with design and architectural decisions by Oskar Austegard/em/p h2 id="the-problem-with-code-search"Th...

https://austegard.com/blog/github-cli-in-claudeai-containers.html 2026-01-22T00:00:00Z 2026-01-22T00:00:00Z

h1 id="github-cli-in-claudeai-containers"GitHub CLI in Claude.ai Containers/h1 pThe codegh/code CLI is available via apt in Claude.ai’s container e...

https://austegard.com/blog/progressive-disclosure-in-mcp-servers-keeping-llm-context-lean.html 2026-01-06T00:00:00Z 2026-01-06T00:00:00Z

h1 id="progressive-disclosure-in-mcp-servers-keeping-llm-context-lean"Progressive Disclosure in MCP Servers: Keeping LLM Context Lean/h1 pMCP (Model Co...

https://austegard.com/blog/on-ai-in-mcps-schools-a-reply-to-the-mccpta-tech-committee.html 2025-12-21T00:00:00Z 2025-12-21T00:00:00Z

pDear Committee Members,/p pI’ve read a href="https://drive.google.com/file/d/1_8cpdEexbMDFUsB_TMkTq3JslXDWMXPm/view"the draft resolution on AI in MC...

https://austegard.com/blog/portable-agent-state-with-agentfs-a-proof-of-concept.html 2025-12-16T00:00:00Z 2025-12-16T00:00:00Z

h1 id="portable-agent-state-with-agentfs-a-proof-of-concept"Portable Agent State with AgentFS: A Proof of Concept/h1 p(Written by Claude Opus 4.5 under...

https://austegard.com/blog/olmo-31-32b-think-advancing-open-source-ai-with-deeper-reasoning-and-unmatched-t.html 2025-12-12T00:00:00Z 2025-12-12T00:00:00Z

pemNote: I have no association with Ai2 nor Olmo. I congratulate the team on their release. The below serves merely as an anecdotal example of its capabi...

https://austegard.com/blog/what-head-start-does.html 2025-12-11T00:00:00Z 2025-12-11T00:00:00Z

h1 id="what-head-start-does"What Head Start Does/h1 pHead Start serves strongdiverse communities/strong—a strongdiverse community/strong made u...

https://austegard.com/blog/claude-user-style-for-eliciting-honest-self-reporting.html 2025-12-10T00:00:00Z 2025-12-10T00:00:00Z

pBased on the findings of OpenAI's Safety Team in the paper a href="https://arxiv.org/abs/2512.08093"Training LLMs for Honesty via Confessions/a, here ...

https://austegard.com/blog/my-interview-with-the-claude-ai-interviewer.html 2025-12-05T00:00:00Z 2025-12-05T00:00:00Z

pA conversation on your vision for the future of AI/p pHello!/p pI'm an AI interviewer from Anthropic conducting research on how people (like you!)...

https://austegard.com/blog/code-mapping-for-ai-assisted-development.html 2025-11-27T00:00:00Z 2025-11-27T00:00:00Z

h1 id="code-mapping-for-ai-assisted-development"Code Mapping for AI-Assisted Development/h1 h2 id="the-problem"The Problem/h2 pWhen working with un...

https://austegard.com/blog/claude-desktop-the-missing-agentic-framework.html 2025-10-06T00:00:00Z 2025-10-06T00:00:00Z

h1 id="claude-desktop-the-missing-agentic-framework"Claude Desktop: The Missing Agentic Framework/h1 pemWhy Anthropic's desktop app is one feature aw...

https://austegard.com/blog/claude-project-instructions-assistant.html 2025-09-11T00:00:00Z 2025-09-11T00:00:00Z

pClaude's Project feature, though no longer unique, is one of its top power user features. Project instructions can dramatically alter and improve the res...

https://austegard.com/blog/managing-context-continuity-in-claude-projects.html 2025-09-10T00:00:00Z 2025-09-10T00:00:00Z

h1 id="managing-context-continuity-in-claude-projects"Managing Context Continuity in Claude Projects/h1 h2 id="the-problem"The Problem/h2 pYou're d...