Skip to content

feat: sync/async per model like Monid, remove health (0.3.3) - #67

Merged
liyangbing merged 2 commits into
mainfrom
codex/sync-run-no-health
Oct 11, 2026
Merged

liyangbing merged 2 commits into
mainfrom
codex/sync-run-no-health

Conversation

@liyangbing

@liyangbing liyangbing commented Oct 11, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Follow the model's sync/async like Monid, and remove the health metrics. Pairs with sandbaseai/sandbase-monorepo#1130 (server: sync models answer in the call; async models return a run id at once).

  • Health removed: HEALTH column in discover, Health line in inspect, Skill §7b.
  • -w never returns an empty result: while a finished media run is still copying its files, run -w / runs get -w keep waiting instead of stopping with no output; a stale "still in progress" hint no longer survives completion; the run id is printed to stderr when waiting starts (so an agent whose shell times out can resume).
  • Codex: managed config adds tool_timeout_sec = 1800, so long sync calls are not cut by Codex's 60s default.
  • Skill 0.3.3: describes the two kinds of tools — sync returns the output directly; async returns a run_id, poll sandbase_run_get every 5–10 s, never re-run. README updated (and the stale www-fallback sentence removed).
  • The MCP bridge is unchanged.

Tests

  • npm test: 175 tests, 174 pass, 1 skip. New: no health in discover/inspect; -w waits through transfer, drops stale hints, prints the run id on stderr; Codex config carries tool_timeout_sec.
  • Live (production via www.sandbase.ai): run sandbase_bfl_flux_3_text_to_image -w printed the run id on stderr and returned the finished image in 48s.

…th (0.3.3)

- bridge: when the server answers running, wait locally (sandbase_run_get
  every 3s, MCP progress notifications) and return the finished result under
  the agent's request id. Long waits for Claude Code and Codex configured
  with a long tool timeout; bounded 50s run_get waits elsewhere. tools/call
  runs concurrently; notifications/cancelled and stdin close stop waits.
- Codex config: tool_timeout_sec = 1800 + SANDBASE_LONG_TOOL_CALLS=1.
- CLI run / runs get wait by default (--no-wait), print the run id on
  stderr, keep waiting through file transfer, drop stale running hints.
- Remove the HEALTH column, Health line and Skill section 7b.
- Skill/README updated; version 0.3.3.
Sync tools return results directly and async tools return a run id to poll,
as the server now does. Revert the bridge's local waiting and concurrency,
and the CLI's wait-by-default. Keep: health removal, -w waiting through file
transfer, no stale running hints, run id on stderr when -w waits, Codex
tool_timeout_sec for long sync calls. Skill 0.3.3 describes the two kinds.
@liyangbing liyangbing changed the title feat: synchronous run results via bridge/CLI, remove health (0.3.3) feat: sync/async per model like Monid, remove health (0.3.3) Oct 11, 2026
@liyangbing
liyangbing merged commit abb7e6e into main Oct 11, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant