An education and learning harness: read-only AI agents that teach, coach, quiz and plan without editing your code.
bun install
bun run dev # start the TUI
./dev.sh # start the TUI with the OpenRouter key from .env (OPENROUTER_KEY=...)
bun run check # lint and type-checkRun tests from a package directory (for example packages/tui), never from the repo root.
The TUI is blocked until the user signs in through Auth0 with a verified @sigmalabs.co.uk email address. Sign-in uses the Auth0 device flow: SigmaBot shows a link and code, the user approves it in their browser, and the session is kept in sigma-auth.json in the state directory. Each launch exchanges the stored refresh token for a fresh, signature-checked ID token.
The Auth0 tenant is defined as SIGMA_AUTH0 in packages/tui/src/sigma-auth.ts and can be overridden with AUTH0_DOMAIN and AUTH0_CLIENT_ID in .env (exported by dev.sh and build.sh). The Auth0 application must be a Native app with the Device Code and Refresh Token grants enabled.
Agents are built into the product (packages/core/src/plugin/agent.ts, prompts in packages/core/src/plugin/agent/), so they are available from any directory. The upstream build, plan, general and explore agents are gone. No agent can edit files, and shell commands that write to the project are denied; the one exception is the updater subagent, which /update uses to finish a conflicted merge and whose shell is limited to the git commands that do so. A project or global config can still add or override agents by ID.
| Agent | Mode | Purpose |
|---|---|---|
tutor (default) |
primary | Explains concepts, guides exercises and debugging with hints, and teaches with the Socratic method. |
it-support |
primary | Fixes the trainee's laptop setup with an unrestricted shell; used by /setup. |
reviewer |
primary | Gives teaching feedback on the learner's code without rewriting it. |
explainer |
subagent | Walks through a file, function or concept line by line. |
quizmaster |
subagent | Generates quizzes and flashcards and grades answers with feedback. |
researcher |
subagent | Finds and summarizes official docs and reliable references. |
updater |
subagent | Resolves the conflicts of a /update merge, keeping the trainee's work, then commits. Only used by /update. |
Slash commands are built in too (packages/core/src/plugin/sigmabot-commands.ts, prompts in packages/core/src/plugin/command/sigmabot/).
| Command | Purpose |
|---|---|
/update |
Pulls the latest course content from SelfPacedPreCourse into the trainee's local fork. SigmaBot adds the repository as the upstream remote if no remote points at it, fetches, and merges its default branch into the checked-out branch with --autostash; the tutor then explains the result. If the merge conflicts, the tutor hands the resolution to the updater subagent. Running /update again while a merge is unfinished reports the remaining conflicts. |
/setup |
Switches the conversation to the it-support agent and works through the course's setup guide: the closest SETUP.md (or INSTALL.md, GETTING_STARTED.md, also in a docs folder) in the current folder or above it, up to the repository root. IT Support skips steps that are already done or meant for another operating system, runs the rest, and hands anything needing a password or personal details back to the trainee. |
/submit |
Switches the conversation to the tutor agent and submits the trainee's work on the current challenge for grading. |
/report |
Files a GitHub issue about a problem with the course content. |
See CONTRIBUTING.md. SigmaBot is built on OpenCode; see LICENSE.