convez.ai
SkillProdutividadeIntermediário
JasonJarvan

CC Dashboard

JasonJarvan/cc-dashboard

Dashboard de ações pendentes em markdown — SSOT para equipes com múltiplos agentes

Ver no GitHub
InicialPluginScore · de 0 a 100

Skill que centraliza ações pendentes de usuários em um único arquivo markdown, servindo como fonte única de verdade para repositórios multi-agente. Funciona como projeção sobre sendbox letters, handoffs e rastreadores externos, permitindo que usuários vejam de relance o que precisam fazer.

Instalação

$npx convez-ai JasonJarvan/cc-dashboard

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
0
Forks
0
Acompanhando
0
Saúde
42%
Último commit
há 4 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Orquestração de múltiplas sessões Claude em repositórios
  • Rastreamento centralizado de ações do usuário sem dependência de SaaS
  • Sincronização de tarefas pendentes entre sendbox, handoffs e trackers externos

Veredito

Solução pragmática para o problema real de rastreamento em workflows multi-agente. A abordagem baseada em markdown puro (git-native) é inteligente; o acoplamento com cc-sendbox deixa clara a intenção de integração profunda em ecossistemas Claude Code maduros.

README

A tradução para o português deste README ainda está na fila.

cc-dashboard

One markdown file is your team's pending-actions board. No SaaS, no SQL, no UI. Any agent session appends rows; the user reads at a glance; anyone observing completion marks done.

A portable convention for a single source of truth dashboard of pending user actions in multi-agent / multi-session repositories. Built as a projection over sendbox letters, handoff briefs, and external tracker tickets — never as a replacement.

The problem

You're running an orchestrator plus N implementer sessions across worktrees. Pending user asks accumulate across seven sendbox/toUser/ letters, two tracker systems (Linear / Multica / Jira), and three markdown state files. The user opens four surfaces and still can't answer "what do I owe right now?".

The fix

One file at docs/Dashboard/index.md containing a markdown table. Each row is one atomic user action — a letter with three independent decisions emits three rows, not one. Any agent session appends rows when writing letters or handoffs. The user reads. Anyone observing completion marks done.

| ID     | Type | Action                                  | Where (detail)              | Blocker for         | Since      | Status |
|--------|------|-----------------------------------------|-----------------------------|---------------------|------------|--------|
| UN-001 | C    | Review + merge MR !49                   | gitlab MR !49               | follow-up L2 ship   | 2026-05-20 | open   |
| UN-005 | A    | Decide: adopt mypy?                     | toUser/lint-policy.md §4.9  | lint-policy-v2 L2   | 2026-05-19 | open   |
| UN-014 | E    | Force-burn `.worktrees/bisect-*`        | manual rm -rf               | disk hygiene        | 2026-05-25 | open   |

Why a markdown file (and not Linear / Notion / Jira / Multica)

  • Same surface as the code. Lives in git alongside the work; diffs in PRs; survives session reset; rebases cleanly.
  • AI sessions append in one line. No API token, no schema migration, no rate limit.
  • Atomic-action granularity. Tracker tools hold L2 work units. cc-dashboard holds the user actions on those L2s — different granularity, different lifecycle.
  • One file beats N files. Sendbox letters fan out (one per recipient role); the dashboard collapses (one per repo). The user scans one table.

What you get

  • Canonical path: docs/Dashboard/index.md (single markdown file, two sections: Active + Archive)
  • Row schema: ID | Type | Action | Where (detail) | Blocker for | Since | Status
  • Six action types with distinct completion signals:
    • A Decision / Approve · B Start session · C Review / Merge
    • D Read + Triage · E Destructive ops · F Ops / Admin
  • Write triggers tied to sendbox letter creation, handoff creation, and external-blocker observation
  • Four mark-done ownership options with default (any session + user)
  • Fourteen-day rolling archive
  • Clean coupling rules with cc-sendbox's sendbox-protocol skill

Full spec: skills/cc-dashboard/SKILL.md.

Install (30 seconds)

git clone https://github.com/JasonJarvan/cc-dashboard.git
mkdir -p ~/.claude/skills
ln -s "$(pwd)/cc-dashboard/skills/cc-dashboard" ~/.claude/skills/cc-dashboard

Claude Code discovers it at next session start. Invoke via the Skill tool with name cc-dashboard.

Adopt in a target repo

cd <target-repo>
mkdir -p docs/Dashboard
# Copy the template from SKILL.md §"File template" into docs/Dashboard/index.md
# Add a thin hook config under your harness conventions (path varies by recipe)
# Cross-link from CLAUDE.md / AGENTS.md "Where to look"

For HarnessStack-managed repos there is an embedding playbook tailored for factory agents: docs/sendbox/toHarnessFactory/from-dashboard-embedding-playbook.md.

When to use

Use when:

  • Three or more active agent sessions accumulate user-blocking asks
  • A single letter typically holds two or more atomic user decisions
  • The user wears multiple roles (driver / courier / session-starter / reviewer) and switches frequently
  • External trackers exist but they hold L2 work units, not user-action granularity
  • Reachability test fails: the user needs four or more surfaces to answer "what do I owe right now?"

Don't use when:

  • Single session, single user, chat-resolved tasks
  • An external tracker provides personal-action granularity and the team uses it consistently

Companion skill

cc-sendbox — defines the underlying letter protocol cc-dashboard projects from. Adopting cc-dashboard without sendbox-protocol works but loses the natural coupling at letter-creation triggers.

Repository structure

cc-dashboard/
├── README.md                                   # you are here
├── LICENSE                                     # MIT
├── skills/
│   └── cc-dashboard/
│       └── SKILL.md                            # main artifact (~250 lines)
└── docs/
    └── sendbox/
        └── toHarnessFactory/
            └── from-dashboard-embedding-playbook.md

Status

  • v0.1 — protocol distilled from the everclaw repository's HarnessStack hook into a portable skill.
  • Per-repo adoption requires a thin hook config declaring language policy and mark-done owner — see SKILL.md §"Adopting in a new repo".

Provenance

Distilled from a research-subagent proposal evaluating Linear / Multica / file-based options for managing multi-agent pending-user-action sprawl. Business-specific identifiers from the source project (EVE-NN issue keys, MR numbers, project-specific letter paths) intentionally absent — only the reusable pattern survives.

License

MIT — see LICENSE.

Ver o original no GitHub

Bom para

Relacionados