Skip to content

In active development · pre-release

All your specs.One tab.

spectant is the local spec companion for developers. One binary reads the Markdown specs in every repository you register and shows where each one stands, what comes next and what needs attention. It comes with a Spec skill for Claude Code, so you drive the work from the AI chat while the app keeps the overview. It runs on your machine, and nothing leaves it.

Watch on GitHubHow it works
curl -fsSL https://spectant.pro/install.sh | shcoming soon

One line for macOS and Linux, once the first release ships.

127.0.0.1:7717live

Dashboard preview

demo data
repos
3
closed
1/6
claims
25/56

~/code/harbor

  • 003Auth flowimplementing
  • 004Billing exporttasked

~/code/lantern

  • 011Offline syncimplementing
  • 012Search indexplanned

~/code/atlas

  • 001App skeletonclosed
  • 002Rate limitsimplementing

Chat + App

Work in the chat. Keep the overview in the app.

spectant comes with a Spec skill for Claude Code. You shape, plan and build your specs right from the AI chat, and the skill writes plain Markdown into your repository. The app runs beside it and shows, live, where every spec stands and what your agents are doing.

In the AI chat

Claude Codelater

/spec-implement 002

  • Reading specs/002-web-console/tasks.md
  • T14 · ISC-334 is takeable, starting
  • Probe green, T14 ticked in tasks.md
  • Next: T15 waits on your answer to T19

The skill does the work

Turn an idea into spec, plan and tasks, then work the tasks with your agent. A handful of commands you can remember, and every result lands as Markdown in your repo.

Beside it, in the app

updated just now
  • 002Web consoleimplementing
  • 003Config loadertasked
  • 004Retention policiesclosed

Waiting on you: T19 · keep the ETag poll?

The app keeps watch

Every spec across all your repositories, what comes next, what waits on you and which agent holds which task. You keep control without reading a single log.

Same Markdown files, no sync in between

Built for Claude Code. The skill follows the open Agent Skills format, so other coding agents that load skills should be able to use it too.

Planned: chat and app, closer together

  • later

    Answer in the app

    Answer an agent’s open question right where the app shows it. spectant writes your answer into the spec as Markdown, and the agent picks it up on its next run.

  • later

    MCP server

    spectant offers specs, next steps and open questions as tools, so your agent answers “what is next in this repo?” from the live model instead of guessing.

  • later

    Open in chat

    Every next step, warning and open question gets a copy action that puts the matching command and its context on your clipboard, as a ready prompt for your chat.

  • later

    Local model panel

    An opt-in side panel with a local model via Ollama, or your own API key if you choose. It answers questions about your specs and never writes.

  • later

    Agent feed

    Hooks report every agent step to the app: task started, probe red or green, question asked. The feed behind the live board.

  • later

    Beyond Claude Code

    The Spec skill for other coding agents that load skills, tested and documented instead of merely assumed.

What it does

Your specs stay in your repos. spectant shows them.

The Markdown files are the single source of truth. spectant reads them, puts them side by side and keeps up while you and your agents work.

  • first release

    Every repo, one screen

    Register your repositories once. Every spec shows up side by side with its stage, its progress and the next open step.

  • first release

    Live, in place

    Edit a spec in your editor and the dashboard follows. No build step, no rendered pages, no refresh.

  • first release

    Local only

    One binary with the UI, fonts and icons built in. The server listens on loopback, and no code path calls out.

  • first release

    Every change traceable

    spectant writes only what you set in the app, like a stage, a review mark or a checkbox. Every change lands in versioned files, so git shows what changed and when.

  • later

    Agent board

    See which agent holds which task in which worktree, and watch its probe go from red to green.

  • later

    Spec skill for Claude Code

    Spec-driven work straight from the AI chat, with a handful of commands you can remember. It writes only Markdown, and the app shows the result.

Sneak peek

design prototype

Where the app is heading.

Screens from the design prototype spectant is being built against, in your theme and with demo data: the app does not render these pages yet.

127.0.0.1:7717
Workspace overview: every registered repository as a card with master and spec claims, specs, warnings and next upWorkspace overview: every registered repository as a card with master and spec claims, specs, warnings and next up
Workspace dashboard: master and spec claims, what is takeable now, next up with the command to run, TL;DR, spec list and warningsWorkspace dashboard: master and spec claims, what is takeable now, next up with the command to run, TL;DR, spec list and warnings
Library: every plan, design, decision and note of a workspace, searchable, with outdated documents flaggedLibrary: every plan, design, decision and note of a workspace, searchable, with outdated documents flagged
Spec dashboard: key numbers, the idea, next step with reason and lanes, beside a rail with what waits on you, warnings and gatesSpec dashboard: key numbers, the idea, next step with reason and lanes, beside a rail with what waits on you, warnings and gates
Status area: why this next step, progress and gates, where the spec standsStatus area: why this next step, progress and gates, where the spec stands
Live board: one lane per column, in flight, waiting and landed, with a needs-you railLive board: one lane per column, in flight, waiting and landed, with a needs-you rail
Tasks tab: checkbox states, lane and state chips, probe mappingTasks tab: checkbox states, lane and state chips, probe mapping
Notes: workspace notepad with anchors to a spec, claim or taskNotes: workspace notepad with anchors to a spec, claim or task

Workspace overview: every registered repository as a card with master and spec claims, specs, warnings and next up

How it works

Three commands. That is the whole setup.

Install once, register the repositories you care about, open the dashboard.

  1. 01

    curl -fsSL https://spectant.pro/install.sh | sh

    Installs a single binary for macOS or Linux, arm64 or x64. No sudo, no shell edits.

  2. 02

    spectant add ~/code/my-project

    Registers a repository. spectant only remembers the path.

  3. 03

    spectant

    Opens the dashboard in your browser, served from 127.0.0.1.

Planned CLI, subject to change until the first release.

Built in public

Where spectant stands today

spectant is built spec by spec, the same way it wants you to work. This is the honest state.

  1. App skeleton and dashboard

    in progress

    The binary, the loopback server and the first dashboard with real repositories.

  2. Shell and spec page

    up next

    A full spec view with status, data, docs and notes in one shell.

  3. Live state and agent board

    planned

    Every stage change as it happens, and which agent works on what.

  4. Spec skill for Claude Code

    planned

    The spec workflow as a Claude Code plugin that shares spectant’s parser, so the chat and the app always agree.

  5. First public release

    planned

    Signed binaries and the one-line install for macOS and Linux.