Concepts
Experts and skills
Every task in Epic Context gets an expert: one per group, each with a method for doing that kind of work well. The method is the same for every client; under it come the organisation's and the project's own rules and learnings.
Any AI can produce a plausible persona or a plausible screen. The difference is whether it works the way a good specialist would, with your facts. Epic Context separates the two: skills hold the method, the graph holds the facts, and the server puts them together for each task.
Eleven skills of two kinds
Three how-to skills, the same for everyone:
| Skill | Answers |
|---|---|
epicontext-core | How does every task run? The workflow (orient, read, write with a reason, connect, land), the hard gates, and which skill to load next |
epicontext-suite | Where does each thing go? The tools of the Epic Suite and the five verbs |
epicontext-context | Is the brain right? Kickoff, filling a project, health checks and repairs, the required connections |
Eight experts, one per group:
| Expert | Group: question |
|---|---|
epicontext-brand | Brand: how do we show up? |
epicontext-strategy | Strategy: where are we going? |
epicontext-users | Users and research: who for and why? |
epicontext-product | Product: what is it? |
epicontext-design | Design: how does it look and feel? |
epicontext-build | Build: how does it work? (frontend and backend) |
epicontext-marketing | Marketing: how do we grow? |
epicontext-performance | Performance: is it working? |
Specialist sub-skills
Under the experts sit sixty-two specialist sub-skills, each one craft or way of thinking: seo, sea and linkedin-ads under Marketing; storytelling under Brand; ogsm and business-model-canvas under Strategy; api-design and web-accessibility under Build; ios-hig and material-3 under Design; workshop-boards for Miro and FigJam under the base. Every expert also has an explore skill to open a question up and a grill skill to stress-test a decision. A briefing names the sub-skill a task needs; list_skills with a query finds the rest.
The task picks the expert
You do not have to tell the AI which expert to use. Each briefing task type names one:
| Task type | Expert |
|---|---|
ui-component, ui-page | Design |
content | Brand |
strategy | Strategy |
research | Users and research |
product | Product |
marketing | Marketing |
performance | Performance |
build-backend | Build |
A general task takes the expert of the block it targets. The write gate checks the match: a block written in a session that never loaded its expert, or under another expert's briefing, is refused in projects that enforce the gate (the default) and flagged in projects set to warn.
Three layers in one call
get_skill("epicontext-design", project_id) returns, in this order:
- The method: the skill itself, the same at every client.
- The organisation's expertise: its expertise document (for Design, the design rules) and the rules and learnings saved for that expert.
- The project's expertise: the same, for this project. A project may differ from its organisation, never from the method's floor.
The skill text stays on the server. A copy on disk goes stale; the plugin for Claude Code only carries pointers that make the agent fetch the current version.
How the expertise grows
save_rulekeeps a choice ("we always…, we never…") as a rule under an expert.log_learningkeeps a finding: what worked, what failed, what a client said.- The editors propose a learning that comes back three times as a rule, and name rules that have no expert.
propose_skill_deltaproposes a change to the method itself. Those are reviewed before they apply, because they reach every organisation.
Your organisation's own skills
An organisation can upload packs of its own skills (an audit checklist, a team's method) under Settings › Organisation › Skills, or an owner or admin loads one from GitHub or skills.sh on the same page (only its texts load, never its scripts). An AI session can preview a link with import_skill, but loading is always a person's act in the web. Every AI that works in the organisation finds them through list_skills and reads them with get_skill, the same way as the shipped skills.
Next
- Briefings and the write gate: how a task gets its context and its expert.
- The editors: the agents that keep the brain in shape.
- MCP tool reference:
list_skills,get_skill,save_rule,log_learning.
Last updated: