The dashboard

Start the dashboard

Use a second terminal in your checkout while a run occupies the first terminal.

e11 dashboard
e11 dashboard --port 4711 --no-open

The default port is 4711. The server binds to loopback. To include other checkouts, register each project and start the shared view.

e11 project add /path/to/another-checkout
e11 dashboard --all-projects

Find your work

The rail lists projects, then intents, then the runs pursuing each story. A run's dot says whether it waits on you: amber pulsing for a landing or an approval, red for a failure, purple for work in flight, green when it is done. The intent's header rolls those up, such as 1 to land, 1 to tidy, or all done. On a narrow screen the rail opens from the menu button in the header.

Read a run page

The page opens with a headline in plain words, such as Artifacts is ready to land, a status badge, the request the work reached on the forge, and one paragraph on how the run went. Under it sits the one thing left for you.

• An accepted run offers Land this change. The page fast-forwards your checkout onto the accepted commit and records the landing, as e11 runs land --merge does. Copy CLI command gives the terminal form instead.
• A landed run offers Release the worktree, as e11 runs cleanup does.
• A running run shows what the worker is doing and its stream. A failed or blocked run shows what failed and the retry or adopt command to copy. A paused run shows the resume command.

Three cards follow: what the run did, what it cost against its ceiling and in tokens, and how many attempts it took.

Act on a run

Actions beside the headline opens everything the run admits right now, in three groups.

• Drive the work carries the engine's own commands: land, release, retry, resume, adopt. Landing and releasing are recorded at once. Retry, resume, adopt and approve need the checkout's execution, which the dashboard is composed without, so the page starts the e11 runs verb itself in the background and follows the run as it moves. Its output is kept under Foundry Home.
• Review approves an item past the checkpoint it stopped at, or sends one back with the reason the next attempt is given verbatim.
• Control the run stops it after the current attempt, or cancels it.

Each asks the same capability the terminal asks. A role that lacks one is refused before anything is touched, with the command to run instead. Elsewhere copies the branch, the run id and every command verbatim, and opens the worktree in your editor.

Read the tabs

• Timeline tells the ledger as a story, one line per thing that happened.
• Report is the report the engine wrote when the run settled, laid out a section per card: what was delivered, what changed, what to do next with its command ready to copy, what went wrong, and what is worth knowing. Money and time are in the cards above rather than in the report. The text matches e11 runs report <run>.
• Changes lists the files the accepted range changed, with each file's patch beside the list.
• Proof lists the controller's own gates, one row per trusted command, with view log for each. The banner names the revision every acceptance check passed at. The worker's own claims sit under it, labelled as claims. The tab appears once a gate has run.
• Raw ledger is every committed row, newest first.

Read the transcript

The worker's own words and calls read as a conversation: what it said as prose, each tool call as a card carrying its family's glyph, its argument, its size and the time it ran. A command shows its output on a terminal surface, an edit shows the lines it took out and put in, a checklist shows what was ticked, a sub-agent shows what it was asked and what it answered. Thinking is folded away behind one line.

Filter by Said, Tools, Edits, Commands, Reads or Thinking, search any of it, and open or close every card at once. The figures beside the search say how many calls, how many edits and how many tokens.

Open technical details

The button under the tabs opens the run id, branch, worktree state, baseline, accepted and landed commits, source path, checkpoint, tokens and report author. The worktree row opens the directory in your editor or copies its path. The branch row links the request the work reached, merged or open, and falls back to a comparison. A request is found by the commits the run accepted or landed first, so it is found whatever branch a person opened it from.

Behind the same door, the evidence set reads as a file tree grouped by attempt. Choose a file to read it: a stream renders as a transcript or as raw numbered lines, a result is pretty-printed, a log keeps its line numbers and wraps on request. A large file is read a chunk at a time.

Read an intent

An intent's page opens with a paragraph on where it stands, then its board.

The board is the board: every column the vocabulary declares, in its order, so a column keeps its place while stories move through it. Each card names its story, says where its latest run stands and what it spent. A card opens the story.

A story opens in a dialog: its runs, the documents about it, and what can be done to it. Start a run plans and drives it as e11 run story <intent>/<id> does. The status select moves its board entry by hand. Full page opens the story at its own address with every document about it rendered.

Read the documents

Documents follow the board: everything the initiative was written in, not only what the artifact store manages. Each is grouped by what it is — brief, PRD, architecture, epics, readiness, stories, specs, reviews, sprint changes, research, retrospectives, deferred work — and read through the presentation its kind deserves rather than as one more rendered page.

The card above each document says what it is before what it says: its kind, its status, the story it is about, when it was written, the checks it records with a tick or a cross, the warnings it carries, and the documents it was written from or is about. The frontmatter its tool wrote is read into those facts rather than shown as text; every field it declares is one door away.

Follow live updates

Views follow store events without polling. Any action from the page refreshes the run, the rail, and the intent's page on completion, and shows a toast. The dot beside live in the header reports live, reconnecting, or lost. Check the serving terminal when the connection stays lost.

Run addresses use /p/<workspace>/runs/<run>. Intent routes also accept /p/<workspace>/intents/<slug>/runs/<run>. Both open the same run page, and ?tab=proof opens a tab. A story has its own address at /p/<workspace>/intents/<slug>/stories/<id>, and ?doc=<path> opens one document of an intent.

Intent routes

Run routes