feat(subagent): add API key pool for parallel subagent execution#2369

Closed

Liewzheng wants to merge 23 commits into

MoonshotAI:mainfrom

Liewzheng:dev-feat-subagent-optimized

Commits on May 27, 2026

feat(subagent): add API key pool for parallel subagent execution
Liewzheng
committed
docs: update env-vars and agents docs for subagent multi-key parallel execution
Liewzheng
committed
docs: update CHANGELOG for subagent multi-key parallel execution
Liewzheng
committed
fix(subagent): only inject key-pool keys for kimi provider; fix TOCTOU race in foreground concurrency check
Liewzheng
committed
fix(subagent): unwrap KeyPoolKimi in isinstance checks; limit key-pool concurrency to kimi; guard prepare_instance failures
Liewzheng
committed
fix(subagent): remove API key from log; use resume model for concurrency cap; restore uv.lock
Liewzheng
committed
fix(subagent): remove API key from builder log; keep timed-out agents as killed
Liewzheng
committed
fix(subagent): address Codex Review round 4 feedback
Liewzheng
committed
fix(subagent): resolve default model and count matching-provider runs for kimi cap
Liewzheng
committed
fix(subagent): clear pre-run timeouts from running_foreground
Liewzheng
committed
fix(subagent): rotate keys for streaming status retries
Liewzheng
committed
refactor(subagent): centralise KeyPoolKimi unwrapping with unwrap_kimi_provider
Liewzheng
committed
feat(subagent): add key health tracking with exponential cooldown to APIKeyPool
Liewzheng
committed
fix(auth): preserve pooled API keys during OAuth refresh
Liewzheng
committed
fix(key-pool): reset key failures when cooldown expires
Liewzheng
committed
fix(key-pool): advance fallback slot when all keys cooling down
Liewzheng
committed
fix(auth): only skip OAuth refresh for wrapped pooled providers
Liewzheng
committed
fix(agent): allow timeout=0 to disable foreground timeout
Liewzheng
committed
fix(agent): reject zero timeout for background agents
Liewzheng
committed
fix(agent): respect resume model overrides when applying provider caps
Liewzheng
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(subagent): add API key pool for parallel subagent execution#2369

feat(subagent): add API key pool for parallel subagent execution#2369
Liewzheng wants to merge 23 commits into
MoonshotAI:mainfrom
Liewzheng:dev-feat-subagent-optimized

Commits on May 27, 2026

feat(subagent): add API key pool for parallel subagent execution

docs: update env-vars and agents docs for subagent multi-key parallel execution

docs: update CHANGELOG for subagent multi-key parallel execution

fix(subagent): only inject key-pool keys for kimi provider; fix TOCTOU race in foreground concurrency check

fix(subagent): unwrap KeyPoolKimi in isinstance checks; limit key-pool concurrency to kimi; guard prepare_instance failures

fix(subagent): remove API key from log; use resume model for concurrency cap; restore uv.lock

fix(subagent): remove API key from builder log; keep timed-out agents as killed

fix(subagent): address Codex Review round 4 feedback

fix(subagent): resolve default model and count matching-provider runs for kimi cap

fix(subagent): clear pre-run timeouts from running_foreground

fix(subagent): rotate keys for streaming status retries

refactor(subagent): centralise KeyPoolKimi unwrapping with unwrap_kimi_provider

feat(subagent): add key health tracking with exponential cooldown to APIKeyPool

fix(auth): preserve pooled API keys during OAuth refresh

fix(key-pool): reset key failures when cooldown expires

fix(key-pool): advance fallback slot when all keys cooling down

fix(auth): only skip OAuth refresh for wrapped pooled providers

fix(agent): allow timeout=0 to disable foreground timeout

fix(agent): reject zero timeout for background agents

fix(agent): respect resume model overrides when applying provider caps

Commits on May 28, 2026

fix(agent): persist resumed model override to store for provider cap counting

fix(agent): defer persisting resume model override until launch succeeds

fix(agent): keep non-Kimi foreground cap global