A collection of task-specific agents. Each one lives in its own folder with its own README, dependencies and tests.
| Agent | What it does | Status |
|---|---|---|
luna/ |
Local-first agentic layer over Jira, Discord and Slack for a student lunar rover team: status intake, standups, meeting action items, blocker/budget alerts, Q&A with citations, design-review drafts, deadline reminders, and a risk manager. | Work in progress, not deployed |
a11y-agent/ |
Visitors report accessibility barriers; an agent triages each report against a WCAG 2.2 success criterion and proposes a patch that a human approves before it deploys. | Work in progress: widget live, agent currently broken (see its README) |
incident-agent/ |
Kubernetes alert investigation with one narrowly-scoped, human-approved action. | Work in progress |
graph-runtime/ |
Checkpointed agent graphs with human-approval interrupts. | Work in progress |
cephlapod/ |
Orchestration harness for coding agents on a pool of local models. | Work in progress, not running |
leviathan/ |
Multi-model news clustering and risk rating. | Work in progress |
Shared rule across these agents: the model only answers typed questions (choose, score, yes/no) with probabilities; code owns the math, control flow and side effects, and anything that writes to another system goes through a human-confirmed proposal.
More agents that aren't published yet, and what each one still needs, are in
CATALOG.md.