zoto-spec-system
v0.7.0 StableDecompose complex initiatives into reviewable specs, judge them independently, then execute with adversarial verification — every deliverable, every subtask.
zoto-agents
Plan and verify your specs. Generate and update your evals. Watch every agent.
Three Cursor plugins, shipped together as peers. zoto-spec-system decomposes complex initiatives into reviewable specs and runs them with adversarial verification. zoto-eval-system stamps static and LLM eval suites side-by-side and updates them with your confirmation when AI primitives change. zoto-cursor-top is a live, htop-style CLI that shows every Cursor agent running on your machine.
Plugins
Independent, composable, free. Install one, install all — they cooperate but none depends on another.
zoto-eval-system
v0.3.1 Stable
Stamp static and LLM eval suites side-by-side, detect drift when AI
primitives change and update evals with your confirmation, and compare
runs across models on a single /canvas chart.
zoto-cursor-top
v0.1.0 Beta
A live, htop-style CLI that lists every Cursor agent on your
machine — IDE windows, cursor-agent sessions, and Cloud Agent
VMs — in one expandable parent-to-subagent tree with scrolling log tails.
How It Works
A short, structured loop per plugin. Run them independently; pair them when a feature has both spec and eval drift.
Spec System — create → judge → execute
1
Create
Run /z-spec-create to decompose a feature into phased
subtasks with dependencies, deliverables, and definitions of done.
2
Judge
Run /z-spec-judge for an independent assessment.
The judge reviews feasibility, risks, and completeness before execution.
3
Execute
Run /z-spec-execute to implement subtasks in order.
Each deliverable is adversarially verified before moving on.
Eval System — create → execute → update / judge / advise / compare
1
Create
Run /z-eval-create to scaffold pytest and
@cursor/sdk backends, generate cases, and stamp the manifest.
2
Execute
Run /z-eval-execute to capture a timestamped run with
static, LLM, and aggregate report.yml outputs.
3
Update / Judge / Advise / Compare
Detect drift, review and approve eval updates per-change, surface
weak graders, recommend the next move, and chart cross-run results
on a /canvas.
Cursor Top — run → expand → tail
1
Run
Run cursor-top (or /zoto-cursor-top inside Cursor)
to discover every IDE window, CLI session, and Cloud Agent VM live.
2
Expand
Drill into any parent agent to reveal its subagents in a collapsible tree, with model, repo, start time, and status per row.
3
Tail
Watch the last few log lines of every agent scroll live, or capture a
snapshot with --once / --json.
Explore
Spec System Quickstart
Install the plugin and ship your first verified spec.
→
Spec System Design
Architecture, agent roles, and the verification model.
→
Spec System Configuration
Full config schema, defaults, and path rules.
→
Eval System Quickstart
Init, configure, and run your first static + LLM suite.
→
Eval System Design
Static vs LLM backends,
_meta.generated, user-confirmation flow.
→
Cursor Top Overview
What the live agent monitor shows and how the hierarchy is built.
→
Cursor Top Quickstart
Install, flags, keyboard shortcuts, and in-Cursor usage.
→
GitHub
Source code, issues, and contribution guide.
→
Commit Specs and Eval Cases Alongside Your Code
Specs land as durable markdown in your repo so the why behind a
change survives the chat session. Eval cases land as runnable pytest and
@cursor/sdk suites so the how-do-we-know-it-works moves with
the code. Pair either with
CRUX Memories
to compound learnings across sessions.
Install
Each plugin installs independently. Add one, all, or none — they cooperate but never gate each other.
cursor plugin install zoto-spec-system
cursor plugin install zoto-eval-system
cursor plugin install zoto-cursor-top