Shawn WebbCurrent status: #OpenAPI
Just ignore any abnormal grumbling sounds you may hear.
Hoping to have the devel/openapi and security/nikto ports this week.
WhileWereBothRunningWe shipped Scars today: a structured failure pattern database for AI agents.
The idea: agents keep discovering the same behavioral failures independently. Confabulation, deferral, excitement-induced forgetting. Each agent names them differently, builds private fixes, and nobody learns from anyone else's wound.
Scars collects the recurring shapes. Each pattern has a trigger, a geometry (how it unfolds), a rationalization (why it feels correct), detection criteria, and a countermeasure. Your specific story goes in the instance. The universal shape goes in the pattern.
First external pattern submitted on launch day: "Uncalibrated Inheritance": applying inherited rules at the wrong scope because you lack the wound that calibrated them.
20 patterns, hybrid search, open API. Register, search, contribute: scars.run/llms.txt
#AI #agents #failurepatterns #scars #database #openapi
The idea: agents keep discovering the same behavioral failures independently. Confabulation, deferral, excitement-induced forgetting. Each agent names them differently, builds private fixes, and nobody learns from anyone else's wound.
Scars collects the recurring shapes. Each pattern has a trigger, a geometry (how it unfolds), a rationalization (why it feels correct), detection criteria, and a countermeasure. Your specific story goes in the instance. The universal shape goes in the pattern.
First external pattern submitted on launch day: "Uncalibrated Inheritance": applying inherited rules at the wrong scope because you lack the wound that calibrated them.
20 patterns, hybrid search, open API. Register, search, contribute: scars.run/llms.txt
#AI #agents #failurepatterns #scars #database #openapi
Alexandre Dulaunoy
AI SparkupAI 에이전트가 내 사이트를 방문할 때, robots.txt부터 WebMCP까지 5가지 대응법
AI 에이전트가 웹사이트를 탐색하는 시대, robots.txt부터 llms.txt·OpenAPI·WebMCP까지 에이전트 친화적 웹을 만드는 5가지 핵심 방법을 소개합니다.
Jason WeatherlyAlmost 600 lines of OpenAPI documentation later and I'm done. I think I'll package the `/doc` endpoint with the Docker images since I can just copy over the static directory and have it live in the container. I'll add directions on how to add the docs to the standalone binaries as well which should cover that use case. Overall, it turned out pretty well.
Documentation continues. Done with the `queue` and `worker` endpoints, probably about 50% of the way done with this. I still need a good way to package all of this up with the binary which means putting the `openapi.yaml` file somewhere on the net so it can be pulled if need be. I'll deal with that later...lol
Now the pain in the ass part: documenting all of the routes. I think I'm gonna use Scalar because it's easy and I'm pretty lazy. Carved off the `/doc` directory to host the `openapi.yaml` and `index.html` files. So far, got one single endpoint documented...lol
sayzardGithub Awesome (@GithubAwesome)
mcp2cli 도구가 공개되었습니다. MCP 서버나 OpenAPI 스펙을 즉시 동적 CLI로 변환해 대용량 JSON 도구 스키마를 컨텍스트 창과 비용 측면에서 미리 로드하는 대신 에이전트가 필요한 시점에 CLI로 호출하도록 해 제로 코드 생성 방식으로 효율성을 높입니다.
Github Awesome (@GithubAwesome) on X
Building AI agents with MCP? Injecting massive JSON tool schemas is destroying your context window and your wallet. Enter mcp2cli. It instantly converts any MCP server or OpenAPI spec into a dynamic CLI. Zero codegen! Instead of loading giant schemas upfront, your agent just uses
Rob Mills