wfdash

A local, read-only browser dashboard over wayfinder maps. It shells out to gh, so it inherits the user's existing auth and no token ever reaches the browser, and it has zero write endpoints — it never mutates the tracker.

Running it

wfdash                      # the overview
wfdash owner/repo#12        # one map
wfdash stop
wfdash restart

If wfdash is not on PATH, this copy came from the Claude Code plugin, which installs no global command. Run the launcher sitting beside this file instead, with the same arguments — Claude Code names that directory ${CLAUDE_SKILL_DIR}:

node "${CLAUDE_SKILL_DIR}/wfdash.mjs" owner/repo#12

A target may be written owner/repo#N, owner/repo/N, or as a GitHub issue URL. A bare invocation opens the overview — the current directory is never used to infer a map, because measured against a 21-map corpus a cwd could unambiguously open only 2 of them.

The command prints the URL and opens a browser at it. Report that URL to the user.

Where there is no browser — a container, a cloud task, a background agent, a machine with no display — it opens nothing and says so on a second line:

http://127.0.0.1:7777/m/owner/repo/12
wfdash: no browser here (no DISPLAY or WAYLAND_DISPLAY) — nothing was opened, the URL above is yours to open.

The dashboard is running either way; only the tab is missing. Pass that line on rather than reporting a browser that did not open. WFDASH_NO_BROWSER=1 forces this, which is the fix for a Mac reached over SSH, where the opener "succeeds" by throwing a tab at whoever is sitting at the console.

What the user sees

• The overview (/) — every wayfinder map they can see, across every repo, as a grid of cards, newest-edited first inside takeable → stalled → uncharted → finished. A live card counts every open ticket by whether it needs the user: 7 hitl · 5 afk, plus · K unmarked when anything is unclassified. A finished map says ✓ all N resolved and an uncharted one no tickets yet. Whether anything can be started is the bar's green segment, not a word.
• The map view (/m/<owner>/<repo>/<number>) — one map as a layered dependency DAG, rank as a column, left to right, with a masthead carrying the destination and a dock carrying prose. Click a ticket for its question, its dependencies by name, and its whole comment thread.

Both pages poll, so claiming a ticket in the terminal shows up without a reload.

Which maps the overview finds

It searches for open issues labelled wayfinder:map owned by whoever gh is authenticated as — read once per process from gh api user, so it is right on the first run without configuration.

user: was measured to silently omit org and collaborator repos. If the user's maps do not all live under their own account, set WFDASH_SEARCH to an expression that reaches them, for example label:"wayfinder:map" state:open org:acme. A search that quietly returns a subset is the one failure this route cannot detect for itself, so an empty or short overview is worth naming to the user rather than reporting as "no maps".

What to tell the user when it fails

The command fails in the terminal for exactly two things, and both are one line to fix:

Everything else — a mistyped repo, an issue that is not a map, a rate limit, a dependency cycle — renders in the page, because the server is fine and only that target is not.

Things worth knowing

• Nothing about the running process is written to disk. No pidfile, no port file. The port is the discovery mechanism: the command probes GET /api/health and reuses the server if it answers {app: "wfdash"}.
• A second invocation opens a second tab, and nothing steers or closes anything. Tab-per-map is the feature.
• The server reaps itself after 30 idle minutes, and an open tab heartbeats /api/health every 10 minutes to stop that happening under it. Those two numbers are coupled and move together.
• Zero npm dependencies and no build step, server and page alike. It needs gh on PATH and authenticated, and Node 18 or newer.

Where this copy came from, and updating it

wfdash ships through two channels, and this file is the same in both.

wfdash install copies this skill into every agent on the machine that loads a SKILL.md and reports what it wrote and what it skipped. It is the user's command to run, not one to run on their behalf.

If the command prints a line saying this installed copy is older than the command itself, the fix is to rerun wfdash install.