$ the-wire · showcase
Goose 1.56 ships subagent controls; Cline fixes dropped prompts and stalled resumes
By RepoJournal · Filed · About Agentic Coding · Composed from the cited sources · methodology
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?v=4)
19 people shipped this
Saoud Rizwan
@saoudrizwan
3 cited
kitlangton
@kitlangton
2 cited
urielefrenvirtusa
@urielefrenvirtusa
1 cited
villahernandez-coder
@villahernandez-coder
1 cited
github-actions[bot]
@github-actions[bot]
1 cited
GitHub Actions (publish-mods)
@actions-user
felixxia-oai
@felixxia-oai
copyberry
@copyberry
ioanbudea-OAI
@ioanbudea-OAI
neriousy
@neriousy
Victor Navarro
@vimtor
rekram1-node
@rekram1-node
jesussamuel-byte
@jesussamuel-byte
elberthc-byte
@elberthc-byte
John-Mason P. Shackelford
@jpshackelford
OpenHands Bot
@all-hands-bot
enyst
@enyst
smolpaws
@smolpaws
Abhijay007
@Abhijay007
Agent tooling got more operator-facing today than contributor-facing: Goose 1.56 adds subagent limits and resume, Cline repaired three ways a working session could silently break, and OpenCode made its CLI flags and memory footprint behave.
v1.56.0 aaif-goose/goose
Goose 1.56 adds automatic thinking-level selection, resuming failed or cancelled delegate subagents, a view-only subagent session link in the UI, and a new GOOSE_MAX_CONCURRENT_SUBAGENTS variable to cap how many subagents run at once. If you run delegated work, the concurrency cap is the setting worth adopting before your next long task.
fix(vscode): keep the first prompt when Plan/Act is toggled while a task starts (#15243) cline/cline
Toggling Plan/Act while a task's session was still starting rebuilt the mode, aborted the turn, and seeded the replacement session before the first prompt had been added, so the prompt was silently dropped and Resume Task picked up without it. Your first prompt now survives the toggle.
fix(desktop): send the resumed transcript to the hub once (#15306) cline/cline
The desktop sidecar sent initialMessages twice, in the session config and top-level to manager.start(), so transcripts crossed the wire twice and sessions over roughly 8 MiB hit the hub's 16 MiB frame limit and could never be resumed. Core only reads the top-level field, so the config copy is gone; large transcripts resume again.
fix(agents): retry a turn whose connection drops after streaming only reasoning (#14795) cline/cline
A connection that died mid-stream after the model had streamed reasoning was treated as a turn with content, so the turn-level retry refused to re-run it. Partial reasoning no longer counts as content for that decision, so these drops retry instead of ending the run.
fix(cli): accept --server and --standalone in any position anomalyco/opencode
The OpenCode CLI only honored --server and --standalone on leaf commands, and commands that connect to a server did not accept them at all, so plugin list or mcp list started the managed service even when you pointed them elsewhere. Both flags now work in any position on those commands.
fix(app,client): wire session transcript retention and evict background transcripts anomalyco/opencode
opencode wired createSessionRetention into packages/app so non-retained session families are evicted via data.session.evict(...), and re-evicts families repopulated by background SSE turn events; the PR cites a 1.74 GB footprint in the live Studio desktop renderer. Worth re-measuring if your desktop client accumulates memory over a long day.
fix(cli): convert Claude hook timeout from seconds to milliseconds during migration google-gemini/gemini-cli
Claude Code hook timeout values are in seconds, Gemini CLI's are in milliseconds, and gemini hooks migrate --from-claude passed them through unchanged, so a configured 30s hook was terminated after 30ms. Re-run the migration for any hooks you already ported, or fix the timeout values in your settings by hand.
feat(cli): add models command and --list-models flag for model discovery google-gemini/gemini-cli
gemini models and gemini models list now enumerate available models, with --output-format json (-o json) for scripted discovery. Useful if you pin model IDs in CI or wrapper scripts instead of reading them off the docs.
Action items
