list: nest spawned lanes, and publish where each lane's chat lives

[JulienMartel]

What

Two things, both about lanes that were spawned from another lane's pane.

• scruff --json gains chat — the checkout whose conversation scruff <name> opens. Equal to path when the lane holds its own chat, the parent's path when it doesn't, and "" when scruff genuinely cannot tell. It reaches all five SDKs and ai/SKILL.md.
• The table nests instead of dropping. A spawned lane is drawn └ <name> under the lane that spawned it, with the marker explained once beneath the table.

Why

A picker wants to hide lanes with no pane of their own — a scruff child checkout has no window, no panel and no transcript, so "opening" one starts a client where no conversation lives. parent cannot answer that: a lane opened with ⌘↵ from inside another lane's pane records the same parentage a scruff child does (new.go:192 and create.go:86 both write the spawning cwd), and that one is a full lane. chat is the honest test, and it is measured rather than inferred.

chat is deliberately not chatHome's answer. resume must always name a directory, so for codex/opencode/pi it falls back to the parent — a good guess when the next step is exec-ing a picker, and a lie when published: every such lane would read "no chat of my own" and disappear from a picker filtering on it. jsonChat publishes an answer only for a client scruff can actually probe, and "" otherwise, which every consumer must read as show it.

The listing nests rather than filters for the opposite reason: a spawned lane carries its own branch and its own PR, in another repo, and closing the parent's pane does not reap it. Once that pane is gone this listing is the only place the branch surfaces. Hiding it everywhere at once is how work goes missing.

Also: chatHome now answers a plain lane (parent == its own main checkout) before the cross-repo compare, which spent two rev-parse calls proving what the parent pointer already said. The listing asks this of every lane.

Verify

make check                      # 200/200 bats, go test, vet, fmt
./scruff                        # spawned lanes indented under their parents
./scruff --json | jq '.lanes[] | {name, path, chat}'

All five SDK suites run green locally (bun, pytest, go test, cargo test, swift test).

Watch out

• Consumer contract: "" means undetermined, never "no chat". A consumer that reads it as falsy hides every lane on a machine that doesn't run Claude. Pinned by list --json: chat is EMPTY, not a guess….
• chat is derived per call and flips from the parent's path to the lane's own the first time an agent leaves a conversation in that checkout. watch will emit one changed event for that — a real change in the lane, not noise, and noted in SPEC §2.2.
• occupied_by is still absent from all five SDK Lane types. Pre-existing, deliberately left alone here.
• The consumer this was built for is haus's Lanes picker (hausfold/haus, same lane name), which degrades safely against any scruff that predates the field.

🤖 Generated with Claude Code

https://claude.ai/code/session_011T9Lb6mUXWEudUcs2E1B9X