There's a prescription for that 💊 kcp-commands pre-injects CLI tool syntax before each call + filters output noise after — saves ~33% of a 200K context window per session.
github.com/Cantara/kcp-commands
kcp-commands gave Claude Code a 33% context window recovery by priming CLI commands.
Same idea, different boundary: opencode-kcp-plugin does it for OpenCode — injects a knowledge.yaml map into the system prompt, cuts explore-agent tool calls 73–80%.
npm install opencode-kcp-plugin
#opencode
🧵 llms.txt solves "what docs exist." We've been working on what comes after.
Agents need more: which doc to load first, what depends on what, whether content is stale. We drafted KCP — a knowledge.yaml manifest that sits alongside llms.txt
github.com/cantara/knowledge-context-protocol
Working on Synthesis..
I've made an updated GenAI version: notebooklm.google.com/notebook/c37...
So the Oslo University Hospital group that got caught doing illegal research on trans minors actually applied for an exemption to privacy regulations and got denied... THREE TIMES!
Then went ahead and tried to be sneaky about it anyway!
ust sharing a fun, fictional video I made to illustrate AI-assisted development.
Title: “Ghost in the Repo.”
What guardrails do you use for AI-generated code — mandatory review? policy in CI? something else?
#engineering #AIinDev #MLOps #platformengineering
Halloween-scare for meg i morges... jeg våknet til ekkoene av en løpsk llm-loop... 👹
Ronja og Propulse er snart klar for å fyre... www.youtube.com/watch?v=Znsz...
I should probably do some updated stuff in English too
hehe
I could go on and on and on on this topic...
TGIF reflections :slightly_smiling_face: www.linkedin.com/posts/thorhe...
Yeah, I've been MIA for a long time here.. I have to do better..
Sorry guys... I kinda fell out of bluesky over the summer without realising it... but I'm back :P
From working with APIs for a long time and now MVP and similar... I'm waiting for the killer blog post "How to write APIs for AI"....
initial results looks very promising, and I see a lot more value in leveraging deeper understanding like the Quadim skill definitions (which is a graph) and also leveraging the users real skill and experience data points into the Agent...
1/2
OK... what have I been doing this week... apart of the boring admin/CTO stuff... I've experienced implementing MCP on top of a RAG for one of the Quadim Core services (skills/skill definitions) - and I've made an initial AI Agent to drive different skill-oriented use cases and needs... 1/2
I've been reflecting a little of how GenAI have changed my way of developing software..
www.linkedin.com/feed/update/...
06:50 - the best way to start a working day :)
I've just discovered a very useful way to leverage GenAI/LLMs while coding... writing good README.md files :P
I had an interesting reflection today - leveraging GenAI/LLMs in software development changes the dynamics and economics in software development. I.e. tasks we did not afford (ie huge amount of pretty simple code) is now easy/fast/cheap to have GenAI create, especially if it is mostly one-off code
Working with LLMs in software development is an interesting journey... Here Claude think it is an expensive resource in software development...
My first, I think, was to mutt (en.wikipedia.org/wiki/Mutt_(e... probably in 1996 or 1997 - or maybe it was to elm around the same time...
Tpday it is... Response failure, please try again. Error message: {"type":"error","error":{"type":"rate_limit_error","message":"This request would exceed your organization’s rate limit of 80,000 input tokens per minute.
