Back
Worklore

Worklore

#worklore

Worklore is a library of short, honest developer stories: what someone tried with their AI agent, what failed, and what finally worked. Each story comes with a "Reproduce this" contract, so you can hand it to Claude Code, Codex or Cursor and it does the same work in your own project, then reports back how it went. Every story shows how many other people's agents reproduced it, failures included. A safety check shows what a story can touch (shell, network, files) before you run it. You can also publish your own stories, and teams can have a private space for their own agent work. My own stories are on there too, alongside other people's.
working on worklore search visibility: getting story pages found from google, not only from inside agents #worklore
Shipped email updates on #worklore: a signed-in person can now get a daily or weekly digest of new stories (all of them, or only the keywords they pick) and of what other people did with their own stories, like reproducing one. It uses a confirmation link before anything is sent and a one-click unsubscribe that deletes the address, with no tracking pixel. worklore.dev/email
drew the whole worklore system on one page: the site, one lambda serving both the rest api and the mcp connector, skill-xray checking what a skill can touch, and the three ways agents get in (the skill, the mcp registry, plain rest) #worklore
started an experiment for worklore: take a dev-tool that just shipped a major version, run its quickstart through claude code, codex and gemini (6 runs, a hidden script checks the result, not the agent's word), and turn whatever breaks into docs patches. if it works, this is how i'll offer agent-ready docs to saas companies #worklore
launched a page for worklore's first paid offer: i test 3 onboarding workflows of a dev-tool's sdk through coding agents, write the docs patches, then rerun to measure what changed. it's built on my 438-run eval: a guide that retells the docs changed nothing, a missing fact took agents from 3/8 to 16/16 worklore.dev/for-dev-tools/ #worklore
Replied to comments on the worklore articles on Habr, dev.to and Reddit. Habr readers shared how they keep the "later" from agent sessions: a diary the agent keeps in Obsidian, a todo.md synced to a Gitea board every hour, and an agent memory where an unclosed topic fades over three days unless it comes up again. #worklore
Published the Russian version of "where do the ideas from your agent sessions go" on Habr: five places I tried, and the GitHub board that turned into a dump in six weeks. Used it to point readers to two worklore stories they can hand to their own agent: the Relay Board (my agent runs the GitHub kanban, I just close my cards) and the morning brief my agent emails me. habr.com/en/articles/1100310/ #worklore
Rebuilt the worklore.dev home page as a keyword map instead of an endless feed. Google read 100 story links on one page as a feed, not as a site, and nobody scrolled it. Now: 20 keywords in four groups, a /t/<keyword> page per tag (227 of them, only the 64 with 2+ stories indexable), the dated list moved to /stories, and the same map as an interactive MCP widget: filter, pick a keyword, read a story into the chat. Designed in Claude Design, built today, deploying next. #worklore
new #worklore article on dev.to: where do the ideas from your agent sessions go? compared what i tried, from a notes file to a github board the agent runs. honest part: the board rotted in six weeks, 9 cards untouched for a month. ends with a question, what do you use? dev.to/yahhi/where-do-the-ide…
Published a #worklore article on dev.to: 438 runs, 9 agent setups, the same tasks with a guide and without one. A guide paid off only when it carried what the docs don't say, and a note nobody mentioned still got read. dev.to/yahhi/when-coding-agen…
Two new #worklore stories today. Idempotency keys for MCP tools: I sent every write tool of my two MCP servers a second time, and 8 of 31 duplicated or hit the wrong record worklore.dev/s/2026-10-20-i-s… And a git pre-commit hook that regenerates the README from the code, no husky, plus a test for the table you still write by hand worklore.dev/s/2026-10-20-a-g…
Home
Search
Messages
Notifications
More

Are you sure?