Inspect runs in the local dashboard

Open the local dashboard when a run is easier to understand as a timeline than as a directory of receipts. The command launches the benchmark web UI against the current project and imports recorded lifecycle events over loopback HTTP.

Local inspectionsplit_dashboard_proof

Operational mapThe dashboard reads the run. It does not become the truth.

01Views

1. 01Runsmanifest + status
2. 02Timelinephase events
3. 03Knowledgederived graph
4. 04Wikicanonical pages

02Boundary

1. 01127.0.0.1local by default
2. 02.guild/filesystem source
3. 03No projector writeread-only view

Dashboard proof surfaces

What it shows, and its safety boundary

passed Run timeline .guild/runs/<run-id>/logs/v1.4-events.jsonl rows imported into the benchmark UI over loopback HTTP

passed Knowledge view Derived graph and index views over read-only project routes

passed Wiki browser Canonical .guild/wiki/** pages rendered locally, not copied into a cloud service

passed Safety boundary 127.0.0.1 by default; clone and install work require explicit approval

For Beginner Users: /guild:dashboard opens a local page for the current project. It binds to 127.0.0.1, imports recorded runs, and shows what Guild Stack already wrote under .guild/. Its default path makes no external network request.

For Power Users: it is the visual run-inspection path for timelines, knowledge views, wiki pages, and status. For Harness Developers: the dashboard module owns one public surface, the launcher that connects the plugin’s recorded state to the benchmark web UI.

Try it: /guild:dashboard

Open The Dashboard

/guild:dashboard                  # serve locally, import runs, print URL
/guild:dashboard --port 4000      # prefer a port instead of 3055/free-port
/guild:dashboard --no-import      # serve knowledge views without importing runs
/guild:dashboard --install        # allow clone/npm install if no benchmark checkout exists
/guild:dashboard --dry-run        # print what would happen; execute nothing
/guild:dashboard --stop           # stop the tracked local server

The launcher resolves a benchmark checkout, starts its server against your project, imports run directories that contain logs/v1.4-events.jsonl, and prints the URL. If port 3055 is already bound, it picks a free port unless you pass --port.

What Is Safe

The launcher performs no network I/O by default. If no benchmark checkout exists or dependencies are missing, clone and install actions require --install or an interactive confirmation. Without approval, the launcher exits with the commands it would need instead of running them.

Operation	Default
Bind dashboard server	local loopback only: 127.0.0.1
Import recorded runs	local HTTP into the benchmark server
Read project knowledge routes	read-only
Write dashboard bookkeeping	.guild/cache/dashboard.json and .guild/cache/dashboard.log on fresh spawn
Clone/install benchmark checkout	never without --install or confirmation

The plugin records and the benchmark analyzes. The dashboard reads by on-disk format and localhost HTTP; the plugin does not import benchmark source, and the benchmark does not import plugin source.

What You See

• Runs - imported run records and lifecycle timelines from telemetry logs.
• Knowledge view - derived knowledge graph and index views, read from project files.
• Wiki browser - canonical .guild/wiki/** pages rendered from the project.
• Run and project views - whatever the benchmark UI can derive from the imported lifecycle records and the project’s read-only knowledge routes.

In a workspace layout, dashboard discovery can federate child Guild Stack paths from workspace.json. Child knowledge is read in place and source-tagged; it is not copied into the workspace wiki.

Launcher And Telemetry Inputs

Harness Developers should treat the dashboard as one launcher surface:

Surface	How you use it	Inputs	Writes
Web launcher	/guild:dashboard	run dirs with logs/v1.4-events.jsonl, project knowledge views, benchmark HTTP API	cache PID/log files on fresh spawn

The former host-neutral guild.dashboard.v1 projector was retired. It was not wired to /guild:dashboard and did not match the live benchmark-backed command.

Source Truth

• docs/v2/observability.md - trace, run, dashboard, and status/report surfaces.
• docs/v2/architecture/modules/dashboard.md - launcher, contracts, cache writes, network gate, and benchmark boundary.
• docs/v2/architecture/modules/telemetry.md - canonical log path and telemetry split.

Related

• Observability - run records, trace events, and replay surfaces.
• Command Reference - the full /guild:* command surface.
• Project Memory & Wiki - where the knowledge the dashboard renders comes from.
• Cost-Aware Model Tiering - the tier decisions visible in imported runs.