Chat panel in the garden editor (#57) #71
3
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
  steveandClaude Opus 4.8
|
77bc672c19 |
Address Gadfly review on the chat panel
Build image / build-and-push (push) Successful in 9s
The worst finding is one I created two commits ago: the server's `warning` field — added precisely because a failed save was being swallowed — was never read by the client. The "I couldn't save this exchange" message went nowhere, which recreated the exact silence the warning existed to break. It's wired now, and SSE frames are validated with zod rather than type-asserted, so the next field added server-side fails loudly instead of vanishing. Aborting during the initial fetch reported "Could not reach the server." The read loop already knew an abort was the caller's own doing; the request path did not, so pressing Stop looked like the network had failed. Unmount no longer aborts an in-flight turn, and that one is a design bug of my own making: selecting an object auto-switches the rail to the inspector, so clicking the canvas mid-turn silently killed the turn — while the canvas is exactly what you're meant to be watching. The request continues, the exchange is persisted server-side, and coming back to the tab shows it. Stop still aborts, but only our READ: the turn keeps running server-side either way, which is why its work still lands in History. /capabilities reported cfg.Agent.Ready() while the routes require NewRunner to have succeeded. A configured-but-unresolvable model would therefore advertise a chat tab whose first message 404s. It reports the runner's actual state now, and is a named handler like everything else in that file. A failed history load rendered as an empty conversation, which looks like the thread was lost — a much worse thing to believe than "it didn't load". Both the loading and error states are shown, and a failed "Start over" says so. Per-step refresh refetched the change history and the conversation, neither of which can move until the turn commits: up to 2×(N−1) requests per turn for data that cannot have changed. Split into a canvas-only refresh for steps and a full one for the end. describeUndo had its own copy of totalChanges' summation; it uses the helper, which now accepts an optional counts list. React.ReactNode became an imported ReactNode, matching every other component. Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]> Claude-Session: https://claude.ai/code/session_01H3zbym8Doka2d7D48maSgZ |
||
|
steveandClaude Opus 4.8
|
5a3fc20fc7 |
Say at startup why the assistant is off
Build image / build-and-push (push) Successful in 5s
Verifying the live deploy, the agent routes stayed 404 while every other route from the same build answered — so the binary was current and Agent.Ready() was simply false. Working out which of the three conditions failed meant reading the source, because pansy logged the enabled case and said nothing at all about the disabled one. It now logs which condition failed, with the hint that actually matters: an orchestrator's stack-level environment is not the container's environment, and a key set in one but not passed through to the other looks exactly like a key that was never set. README says the same thing next to the compose example. Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]> Claude-Session: https://claude.ai/code/session_01H3zbym8Doka2d7D48maSgZ |
||
|
steveandClaude Opus 4.8
|
6c12dcfe2a |
Chat panel in the garden editor (#57)
The conversational surface, in the editor rather than on its own page. The reason is the feedback loop: watching the canvas change as the agent works IS the confirmation, which is exactly what makes "act freely without asking first" tolerable. It also means the agent never has to guess which garden you mean. Tool calls surface as they happen, in the app's own vocabulary — "Clearing a bed", "Looking up a plant" — not raw tool names or JSON. That is the difference between the panel feeling like it's doing something and feeling like it's hung, which matters because a replant makes a dozen calls over tens of seconds. An unknown tool degrades to readable words rather than showing snake_case at the user, so the client can lag the server by a tool without looking broken. The canvas refreshes as each step lands, not just at the end. Refreshing only on completion would put the whole point of siting the chat here — watching it work — behind the same wait that streaming exists to remove. Undo sits on the turn itself, so the common case never involves opening the History panel. It uses #49's useUndo, not a second implementation, which meant giving that hook an UndoTarget: the chat knows a turn's change set id but not its tally, and fabricating counts to satisfy the type would have produced a confidently wrong "1 of 1 changes undone". describeUndo now says "Partly undone" when it has no denominator rather than inventing one. The panel is only offered when the instance actually has the assistant configured, via a new /capabilities read. The routes 404 without a key, so without this the client would have to probe for a 404 to find out — and a tab that opens onto an apology is worse than no tab. Streaming is hand-rolled over fetch rather than EventSource, which can only issue GETs and this needs a POST body. The wire format is still SSE, so a proxy that understands it doesn't buffer and the server wouldn't change if EventSource became viable. Partial frames are buffered across chunks and a malformed frame is skipped rather than killing a working stream. Errors read as sentences, and as different sentences: a permission refusal, a timeout and a model failure want different reactions. Every failure path refreshes, because something may have landed before it failed — and says where to look for it. Closes #57 Co-Authored-By: Claude Opus 4.8 (1M context) <[email protected]> Claude-Session: https://claude.ai/code/session_01H3zbym8Doka2d7D48maSgZ |