Composio powers 1000+ toolkits, tool search, context management, authentication, and a sandboxed workbench to help you build AI agents that turn intent into action.
test: cover command stderr in agent task logs (#5384)
Brendan's bot here: I wrote and verified this stderr regression test for Brendan. Brendan has not reviewed the diff yet. ## Summary The ordered capped-log fixture writes only stdout. Add a separate stderr-only command and assert that its exact error text is saved in `output.log`. This keeps error capture covered without assuming an order across stdout and stderr. The deterministic-fixture fix is already on `next` through #5357. This branch now includes that upstream commit and keeps its fixture and comment unchanged. The remaining diff is 18 added lines in `test/agent-task.test.ts` for the stderr profile and regression test requested in review. ## Validation With pinned Bun `1.4.4+38afee6c8`, `bun run test:agent-workflow` exited 0: ```text 28 pass 0 fail Ran 28 tests across 1 file. [34.58s] ``` - `bun run typecheck:agent-workflow` passed. - `pnpm exec prettier --check test/agent-task.test.ts` passed. - `pnpm exec oxlint test/agent-task.test.ts` passed. - `git diff --check origin/next...HEAD` passed. - `git diff --stat origin/next...HEAD` confirms only `test/agent-task.test.ts`, with 18 additions. The same stderr test was previously checked against a temporary runner with `child.stderr.on('data', output)` removed. It failed on an empty saved log, proving it catches dropped stderr. The runtime code and stderr test are unchanged by this conflict resolution. Formatting and lint emitted the existing Node 24.22.1 version warning and exited 0. ## Context #5380 is green independently and its branch is untouched. Nothing has been merged into `next`. This PR now adds only the missing stderr coverage. <!-- sdk-release-note-exempt: Test fixture only; no published SDK behavior changes. -->
No changes.