dashboard

treeship dashboard starts a read-only localhost control plane over local Treeship instances and their sealed .treeship session packages. It is the browser companion to treeship ui: the terminal UI is for live terminal operation, while the browser dashboard is for reviewing Treeships, attached agents, receipts, reports, trust status, and workspace capabilities.

Usage

treeship dashboard
treeship dashboard --port 55339
treeship dashboard ssn_42e740bd9eb238f6
treeship dashboard --root ../another-project --root /srv/agents/.treeship

Arguments

Options

What It Shows

The dashboard reads from .treeship/sessions/*.treeship/ for the current workspace and any extra --root workspaces, then renders:

• The current Treeship instance, including ship_id, local root, Hub status, receipt count, report count, proof package count, and attached agents
• Trust status with local verification, review load, agent attribution, and capability readiness
• A receipt index with local verification status
• The latest report summary
• An agent work map showing which agents did work across receipts
• Collaboration and handoff edges from receipt agent graphs
• Receipt anatomy: report, canonical receipt.json, Merkle proof data, and local verification
• A review queue for warnings, failed checks, failed commands, sensitive reads, network calls, and missing capture signals
• Treeship capability status for daemon evidence, MCP/tool capture, model/token capture, Hub sharing, and local verification
• Raw receipt.json links for automation
• Receipt reports rendered from the current embedded template

Control-Plane Shape

The dashboard follows a local-first control-plane model:

• The CLI process serves the browser UI and API from one local process.
• Local .treeship packages remain the source of truth.
• The dashboard is a client of the same JSON data that scripts and agents can consume.
• Commands shown in the UI are copyable so a builder can leave the browser and reproduce the same action in a terminal.
• Remote sharing through Hub is optional; local verification is still the trust anchor.

Review Queue

The review queue is the daily operator surface. It answers: "What do I need to inspect before I trust or share this agent run?"

Rows can include:

• Package verification failures or warnings
• Failed commands
• Sensitive file reads
• Network connections
• Ports opened
• Missing model/token capture

When a row has a concrete fix, the dashboard shows a copyable command such as:

treeship package verify .treeship/sessions/ssn_42e740bd9eb238f6.treeship
treeship daemon start
treeship add
treeship hub attach

Agent Work

The agent work map answers: "Which agent did what?"

It aggregates each captured agent identity across sealed receipts inside the current Treeship and shows:

• Latest receipt link
• Which receipts/reports the agent appears in
• Sessions seen
• Action count
• Files read and written
• Commands run
• Network or review signals
• Captured model information when present

Collaboration

The collaboration view answers: "Which agents worked with which?"

It reads agent_graph.edges from receipts and surfaces parent-child spawns, handoffs, returns, and collaboration edges. If no edges are shown, the current receipts did not capture multi-agent relationship events.

Receipt Anatomy

Receipt anatomy answers: "What is the full Treeship receipt and report?"

It maps the selected run into four layers:

• Report: the human-readable summary
• Receipt JSON: the canonical package record
• Merkle: leaf count, inclusion proofs, and root
• Verify: local package checks and signature count

Treeship Capabilities

Treeship capabilities answer: "How well is this workspace instrumented?"

It checks whether:

• Sealed receipts exist
• Local package verification is passing
• Daemon evidence appears in receipts
• Agent or MCP tool events appear in receipts
• Model/token usage is captured
• Hub sharing is configured

Data Source

The dashboard is local-first:

• It binds to localhost.
• It reads sealed .treeship packages from disk.
• It serves package files read-only.
• It does not make Hub the trust root.

The dashboard exposes read-only local JSON endpoints for scripts, agents, IDEs, and future UI clients:

Relationship to preview.html

Every .treeship package can contain a generated preview.html. The dashboard renders receipt reports from receipt.json with the current embedded template, so older packages still benefit from the latest report design when viewed locally.