Skip to main content
← Back to list
01Issue
FeatureShippedSwamp CLIPublic
Assigneesskunk-ape

Relationships

#4751 Swampify: start offers windows up to all history, you choose where your agent opens, and start and the dashboard share one breadcrumb

Opened by skunk-ape · 11/14/2026· Shipped 11/14/2026

Part of Swampify onboarding. Follows #4705 and #4713, from an external tester's first run.

Repo: swamp (swamp start page, the Swampify page, the start briefs and start's launchers only).

Goal

Three changes:

  1. "Start from your past sessions" can look back further than 90 days.
  2. The user chooses where their agent opens, and is told plainly where it went: a new terminal window, herdr, T3 Code, or their own agent app via a copied prompt.
  3. A consistent breadcrumb between start and the dashboard, and a Swampify launch leads to the Swampify page.

No new CLI command or flag.

1. Longer windows

The tester keeps far more than 90 days of history, and the page offers only 7, 30 or 90 days. SwampifyKickoff already accepts up to 365.

  • Offer 7, 30, 90, 180 and 365 days, and all history. Keep 30 as the default.
  • All history is a window with no start date. The kick-off text says so in its window line (for example Window: all history), and the skill reads it as "no since". assess already reads everything when since is left out.
  • That changes the kick-off text #4705 pinned. Agree the exact window line with #4752, which teaches the skill to read it, and pin it with a test on both sides.
  • Next to the longer choices, a short note says a longer window takes longer to assess. The skill still says how many sessions it will read and asks first.

2. Where the agent opens

Today start opens a terminal window (macos_terminal, windows_terminal, linux_terminal) or, when it can't, shows a terminal command worded as an error (copy_command). The tester works in agent apps (herdr, T3 Code, Zed), saw "can't open a terminal" on Swampify!, and would rather have used those.

Targets

Add two launch methods, offered only when detected, and make the copied prompt a normal choice:

  • herdr (a terminal workspace manager for agents).
    • Detected when herdr is on PATH and herdr status reports a running server. With no server there is no herdr window to open in, so it isn't offered.
    • Launch: herdr workspace create --cwd <workspace> --label <name> --focus, then run start's existing launch command in its pane (herdr pane run, or herdr agent start and herdr agent prompt). Confirm the reply shapes with herdr api schema --json.
  • T3 Code (an app that drives Claude Code, Codex and other agents).
    • Launch: t3 project add <workspace> --title <name>, then t3 thread start <workspace> "<launch prompt>" --provider <id> --model <slug> --json with the agent chosen on the start page, then t3 app <workspace> to show it.
    • Detection: t3 on PATH and its server running. t3 status --json doesn't test the connection, and t3 thread list exits 0 with "No running T3 Code server" when none is up, and runs T3's migrations. Find a cheap read-only signal (the server's lock or port file) rather than calling thread list on every page load.
    • Check: the provider ids for Claude Code and Codex, that a Claude thread in a workspace loads .claude/skills/swampify (the brief's read-by-path line covers it if not), and whether its transcript lands under ~/.claude/projects.
  • Copy the prompt ("Use my own agent"). Saves the brief and copies one prompt that works in any agent: read the brief at its path and follow it, and read the swampify skill at .claude/skills/swampify/SKILL.md if it isn't loaded. Opens nothing.
  • New terminal window, as today.

Detection is a cheap check like start's other signals: PATH, a running server, nothing read.

User control and notice

  • Choose. The launch step and Swampify! show "Open in" with the detected targets, the terminal window when one can be opened, and Copy the prompt. The default is the target the user picked last, read from the latest start session's launch record (no new storage); with none, the first detected of herdr, T3 Code, terminal window.
  • Before. The button says where it will open, for example "Open in herdr" or "Open in T3 Code", and the agent it will run.
  • After. Say where it went and how to get there: the herdr workspace label, the T3 project and thread, the terminal window, or "prompt copied, paste it into your agent". The launch record keeps the method.
  • When a target fails. Say what failed in one line and offer the other targets. A missing terminal is a fact, not an alert.
  • The same choice and notices apply to the kick-off launch on the start page and to Swampify! on the Swampify page.
  • Copy and layout can be iterated in the session.

3. Getting from start to the dashboard and back

The tester couldn't find their way from the start page to the Swampify page. On the resume screen with the copy-command steps there was no dashboard link at all. Elsewhere the only links are a small "Open dashboard →" chip and an "Open the dashboard" button shown only after a window opened, and both go to the workspace overview. The dashboard has no way back to start.

  • One breadcrumb everywhere. Every start screen with a workspace open (home, task, agent, trust, resume, launched, and the copy-command and failure states) shows the same breadcrumb: swamp start › workspace › Dashboard, with Swampify added once the workspace has a candidate set. The dashboard, when served by the start server, shows the matching trail back to start.
  • From the Swampify flow, a direct link. After a kick-off or a Swampify! launch, whatever the target did (opened, copied, failed), the page says to review the offer on the Swampify page and links straight to /dashboard/w/<workspace>/swampify. The resume screen of a workspace with a candidate set shows the same link.
  • Then redirect. After a Swampify-flow launch that opened the agent, go to the Swampify page after about 15 seconds, with a visible countdown and a "Stay here" control that cancels it. When the prompt or command was copied, start the countdown only once it has been copied. A failed launch never redirects.
  • The Swampify page shows nothing new until the agent writes the offer; it refetches when the tab comes back into view (#4713).

Out of scope

  • How the skill reads the window line (#4752).
  • Support for reading other agents' history (#4752).
  • Other agent apps beyond herdr and T3 Code; they use Copy the prompt.

Done when

  • Unit tests cover the window choices, all history in the kick-off text, and the bounds SwampifyKickoff still enforces.
  • Unit tests cover detection for herdr and T3 Code (installed or not, server running or not), with mocked commands.
  • Unit tests cover each launcher's command sequence, its outcome and failure, the copied prompt (names the brief and the skill path, contains no other user input, opens nothing), and the default target from the last launch record.
  • Unit tests cover the breadcrumb on every start screen and its Swampify entry, the trail back to start on the dashboard, the direct Swampify link in each launch outcome, and the redirect: countdown, Stay here, starting only after a copy, and no redirect after a failure.
  • An integration test through the start server saves a kick-off with all history and launches it through a fake herdr and a fake T3, checking the brief, the window line against the pinned wording, and the launch record.
  • A manual run on a machine with herdr and T3 Code: a kick-off opens in each, and Swampify! does too; with neither running, Copy the prompt and the terminal window still work.
02Bog Flow
✓OPEN✓TRIAGED✓IN PROGRESS✓SHIPPED+ 1 MOREASSIGNED+ 5 MOREREVIEW+ 16 MOREPR_LINKED+ 2 MORESESSION_SUMMARIZED

Shipped

11/14/2026, 8:29:31 PM

Click a lifecycle step above to view its details.

03Sludge Pulse
skunk-ape assigned skunk-ape11/14/2026, 4:58:44 PM

Sign in to post a ripple.