skydive-cli 0.6.0-beta.39 → 0.6.0-beta.41

package/CHANGELOG.md

@@ -9,6 +9,8 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Changed
 
+- `skydive secrets` and `skydive keys` now accept `--agent <id or name>` (the same flag every other agent-targeting command uses), resolving the name against your agent roster. They used to take only a raw `--agent-id`, so passing the name every other command accepts 404'd "agent not found" for a perfectly real agent; `--agent-id` still works.
+
 - `/fork` in the chat TUI now works while the agent is mid-run. It used to refuse with "finish or cancel the run first"; now it copies the conversation up to the last completed message and opens the copy, while the agent keeps working on the original — reopen the original to reattach, exactly as after `esc`. This matches the web "Fork chat" menu item, which never had the restriction.
 - The two cards pinned above the chat TUI composer — **subagents** and **plan** — can now be minimized to a single header row. Click a card's header (the `▾`/`▸` chevron shows which way it goes), or use `/subagent minimize` and `/plan minimize`; the same again, or `/plan expand`, brings the rows back. Minimized, a card keeps its label and its progress counter (`Subagents 7/10`), so a ten-task fan-out costs one row instead of ten and the rest of the terminal goes back to the transcript. `/plan expand` and `/plan collapse` keep their meaning (unfold and re-fold the completed rows the card windows away); clicking the card's _body_ still does the same, and the hints now say which surface does which.
 - Reopening a conversation whose subagents and plan are all finished now opens both pinned cards minimized to their header row (`▸ ✔ Subagents 4/4`, `▸ Plan 5/5`) instead of spending up to ten rows above the composer on finished work. A card that opens this way pops back open by itself the moment new work goes live in that conversation (a fresh fan-out, a new in-progress step); a card you minimized yourself stays minimized, as before. Work still running when you open the conversation shows its rows, exactly as it did.
@@ -19,6 +21,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ### Added
 
+- `skydive agents git token` — create, list, and revoke git-proxy tokens for an agent's repo from your machine. `agents git token create <agent>` prints a read_write token plus the tokenless remote URL (the same routes the web settings page uses); `… git token list <agent>` shows the agent's tokens by id, and `… git token revoke <agent> <token-id>` revokes one. The command's job ends at the credential: it writes nothing to git on your machine, so what to do with the token (credential store, remote URLs) stays with whatever tooling you already pipe credentials through. `auth login` now points at it for laptop git setup.
 - Files an agent shares into the conversation now show as a saveable row in the chat TUI. When the agent sends you a file (a report, an export, a screenshot), it appears as a `📎` row with the filename and size and a "click to save" hint; clicking it downloads the file to your `~/Downloads` (falling back to your home directory) and opens it with your default app, then the row shows where it landed. Clicking a saved row again re-opens it. Previously these files were only reachable in the web app — in the terminal they were invisible.
 - `platform ask` forms now render in the chat TUI. The card lists each question with lettered options (A/B/C/D plus Other); `ctrl+r` opens a full overlay picker (one question at a time, arrows + enter, Other as a typed row — the Claude Code interaction). The answer sends as a normal message. Headless `chat -p` prints the question as an `[action needed]` block instead of dropping it.
 - Every API request now reports this install's version (`x-skydive-cli-version`) so the server can refuse features the installed CLI cannot render — `platform ask` needs the #21411 canary (`0.5.0-beta.43` or newer), not 0.4.0 and not earlier 0.5.0 betas.