Baseline Core
Open-source skills system that wires your business into AI
101 followers
Open-source skills system that wires your business into AI
101 followers
Baseline Core is an open-source skills system for AI agents. Run one command and your AI tools can research markets, write PRDs, plan sprints, and design user flows -- all grounded in your business context. 12 skills, 14 frameworks, 34 reference files. Works with Claude Code, Cursor, Codex, ChatGPT, Gemini, Windsurf, and GitHub Copilot. Free forever.
Hey everyone! I'm Trent, founder of Baseline Studio. I research how product teams can work better with AI.
Baseline Core is a free, open-source skills system that wires your business context into any AI tool. You run one command, answer a few questions about your business, and your AI agents can start doing real product work -- research, PRDs, sprint planning, user flows, design, prototyping, positioning, and more.
It works with Claude Code, Cursor, Codex, ChatGPT, Gemini, Windsurf, and GitHub Copilot. Tool-agnostic by design.
I built this because I kept seeing founders and small teams either skip product work entirely or get generic AI output that didn't sound like their business. Baseline fixes that.
Would love to hear what you think. Happy to answer any questions!
@baseline_studio On the skills side, what does a Baseline Core skill look like under the hood, is ux-design just markdown in skills/, or can a skill run code too? The npx @baseline-studio/cli init scaffold that drops skills/, context/, frameworks/, and AGENTS.md feels like the right path for tool-agnostic use across Claude Code, Cursor, Codex, and Copilot. The trust builder for teams is drift control, version skills plus core/voice.md, stamp outputs with the versions used, and ship a tiny eval set so a tweak doesn't silently change PRDs or sprint plans.
@baseline @piroune_balachandran Great question. A skill is a markdown file with a YAML manifest. The manifest tells the AI which context and reference files to load when the skill runs. So it's not just one file, it's a skill file plus supporting references that all load together. Everything is markdown, no code execution. The AI tools themselves can run code through their own capabilities, but Baseline is purely the knowledge and methodology layer.
The drift control idea is really interesting. The system is versioned at the repo level right now, but stamping outputs with skill versions is something I want to explore. Appreciate the feedback.
@baseline_studio Trent, how flexible is the configuration regarding what exactly gets fed into the context? For example, if I'm making a skill for DB refactoring, can I feed it only the database schema and migrations, ignoring the marketing docs? I want to keep the focus narrow so the model doesn't get distracted or start hallucinating
@baseline_studio Hey Trent. Congratulations on the launch! Are skills represented as prompts, workflows, or state machines? Can skills encode decision logic and thresholds?
The markdown-as-skill pattern is underrated — keeping everything as plain files means it stays portable and version-controllable without any lock-in. What I'm curious about: when you say the skill loads supporting context files together, does the agent get all 34 reference files at once or does it selectively pull based on the specific skill being run? Managing context window size feels like the main tradeoff here vs. just dumping everything in.
@cogotemartinez Each skill has a manifest.yaml that declares exactly which context, frameworks, and reference files it needs. Reference files are nested inside each skill's folder, so the UX design skill has its own set of 6 references while research synthesis has a different set. The agent reads the manifest first and only loads what that specific skill requires. Not all 34 at once. Agents[dot]md is what connects everything. It routes tasks to the right skill, tells the agent where to find the manifest, and defines the execution protocol. The workflow orchestration framework also handles context management across sessions. For larger tasks it scopes work into multiple sessions and generates a plan so context stays clean and output quality doesn't degrade. That said, managing context within longer sessions isn't always consistent yet and it's something I'm actively looking for better solutions to.
Trent showed me this system about a month ago and I've been experimenting with it since then; and my honest take is, this is a great unlock! With this system I have been able to supercharge my claude-code usage (tbh I also got into claude-code because of this system) to the extent where I am effectively only reviewing PRs now! Another thing this has really done for me is unblock me on all the non-technical tasks (like voice and communication or customer profile and positioning) and concretely articulated my ideas and documented them! I am now adapting the @Baseline Core system to fit my exact workflow and exploring how I can build without any employees but with the impact of a team of ten.. That is what this system truly unlocks!!
@isthispalash This is incredible feedback. The fact that you've gone from experimenting with it to adapting it to your own workflow is exactly what the system is designed for. The context layer is what makes that possible. Once it knows your business, every skill gets more specific to how you work. The "impact of a team of ten" framing is something I think about constantly, that's the real unlock here. Not replacing people, but giving a solo founder or small team the structured output that used to require dedicated roles across product, design, marketing, and strategy. Really appreciate you sharing this!
The thing I keep running into with AI coding tools is the context problem - every new conversation starts cold, no memory of past product decisions or constraints. Makes it feel like you're constantly re-explaining your own product. Does Baseline Core persist context across sessions, or is it more of a 'load once at session start' thing?
@mykola_kondratiuk Context doesn't persist automatically across sessions, but the system handles this in two ways. First, for best results you work out of the Baseline Core repo so every session has your full context available. Your business identity, voice, users, competitive landscape, all of it loads through the manifest when a skill runs. So you're not re-explaining your product, the context files do that for you.
Second, for larger work that spans multiple sessions, the system has a session planning framework that scopes deliverables across sessions, defines what context to load in each one, and generates a project plan with handoff notes so continuity survives between sessions.
Also worth noting, Claude just rolled out Auto Memory which does persist learnings across sessions. That pairs really well with Baseline Core since the system handles structured context and methodology while Auto Memory handles the session-to-session continuity.
This is awesome! Does it integrate with OpenClaw? How did you curate the skills?
@dan__cleary Thank you! I haven't used OpenClaw yet, but Baseline is modular by design. The skills, frameworks, and context files are all standalone markdown, so in theory they should be portable into other systems but I'd love to explore that more!
As for how I curated the skills, they come from my experience across the full product development spectrum. Each one maps to a phase of pre-engineering product work that I've done hands-on and refined across client engagements.