fix(providers): adapt to claude-agent-sdk 0.2.121 native-binary auto-resolve

[dsorensen]

Summary

• Bumps @anthropic-ai/claude-agent-sdk from ^0.2.74 → ^0.2.121.
• Drops the dev-mode short-circuit in resolveClaudeBinaryPath so dev follows the same env → config → autodetect chain as binary mode.
• The native-binary detection in shouldPassNoEnvFile now sees the real resolved path and correctly suppresses --no-env-file for native installs.

Why

claude-agent-sdk 0.2.121 stopped shipping a bundled cli.js and now resolves to a native platform binary by default. Dev mode used to return undefined and let the SDK auto-resolve — but with 0.2.121 the SDK auto-resolves to the native binary, and shouldPassNoEnvFile(undefined) then returns true, so --no-env-file is forwarded to a native binary that rejects unknown flags. Result: every Claude workflow node fails with exit code 1 immediately on spawn.

Reproduced in three back-to-back ai_demo workflow runs and confirmed against direct sendQuery repro (#878 in session memory).

Test plan

• bun run validate passes locally (type-check, lint, format, all package test suites).
• binary-resolver-dev.test.ts rewritten as a regression check that dev mode honors env / config (no short-circuit). New assertions are mode-agnostic.
• provider.test.ts mocks ./binary-resolver to return undefined so existing executableArgs: ['--no-env-file'] assertions don't depend on whether the test host has a native Claude install at ~/.local/bin/claude.
• Manual repro: re-running archon workflow run claude-plan-full-sdlc against any of the three ai_demo projects no longer hits the --no-env-file crash on the first node spawn.

Notes

• Tracked separately: implement-loop signal-mismatch surfaced in those same three runs is filed as claude-plan-full-sdlc implement loop never exits — agents complete work semantically but never emit ALL_TASKS_COMPLETE token #1471 (workflow-engine bug, not a provider bug).

🤖 Generated with Claude Code

Summary by CodeRabbit

• Dependencies

  • Updated Claude SDK dependency to version 0.2.121.

• Improvements

  • Claude native binary resolution now consistently applies environment variable and configuration overrides across all execution modes, ensuring predictable behavior.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: f02cd89e-9d41-409d-8aae-7391f35ecd77

📥 Commits

Reviewing files that changed from the base of the PR and between 7d06773 and 33ef830.

⛔ Files ignored due to path filters (1)

• bun.lock is excluded by !**/*.lock

📒 Files selected for processing (4)

• package.json
• packages/providers/src/claude/binary-resolver-dev.test.ts
• packages/providers/src/claude/binary-resolver.ts
• packages/providers/src/claude/provider.test.ts

---

📝 Walkthrough

Walkthrough

The Claude binary resolver no longer applies dev-mode special handling, now consistently following an environment-variable-first resolution chain across all modes. Dependency on @anthropic-ai/claude-agent-sdk upgraded from ^0.2.74 to ^0.2.121. Related tests refactored to verify the unified behavior and eliminate environment-dependent test results.

Changes

Cohort / File(s)	Summary
Dependency Upgrade package.json	Updated @anthropic-ai/claude-agent-sdk from ^0.2.74 to ^0.2.121.
Binary Resolver Logic packages/providers/src/claude/binary-resolver.ts	Removed BUNDLED_IS_BINARY conditional gate; resolveClaudeBinaryPath now consistently applies env/config/autodetect resolution chain in all modes instead of short-circuiting to undefined in dev mode.
Resolver Tests packages/providers/src/claude/binary-resolver-dev.test.ts, packages/providers/src/claude/provider.test.ts	Refactored dev-mode tests to validate proper resolution chain using fileExists spy; added env variable teardown; mocked binary resolver in provider tests to eliminate environment-dependent behavior.

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~22 minutes

Possibly related PRs

• fix(providers): replace Claude SDK embed with explicit binary-path resolver #1217: Directly modifies Claude binary-resolution logic and tests in the same resolver files.
• feat(providers): autodetect canonical binary install paths for Claude and Codex #1361: Modifies binary-resolver.ts and binary-resolver.test.ts with changes to path resolution behavior.
• chore(deps): bump claude-agent-sdk to 0.2.121, codex-sdk to 0.125.0 #1460: Upgrades the same @anthropic-ai/claude-agent-sdk dependency to ^0.2.121.

Poem

🐰 Hoppy paths now run on truth,
No shortcuts in the binary booth!
Dev-mode checked, now gone for good,
One clean path, just as it should. ✨
Tests behave, no secrets hid,
The resolver's heart is vivid!

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Description check	⚠️ Warning	The description covers the core problem, why it matters, what changed, and verification approach, but omits multiple required template sections (UX Journey, Architecture Diagram, Label Snapshot, Change Metadata, Validation Evidence, Security Impact, Compatibility, Human Verification, Side Effects, Rollback Plan, Risks and Mitigations).	Complete the description template by adding the missing sections: UX Journey, Architecture Diagram, Label Snapshot, full Validation Evidence, Security Impact, Compatibility details, Human Verification checklist, Side Effects assessment, Rollback Plan, and Risks/Mitigations.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and concisely describes the main change: adapting providers to a new SDK version's native-binary auto-resolve behavior by updating dependency and resolver logic.
Docstring Coverage	✅ Passed	Docstring coverage is 100.00% which is sufficient. The required threshold is 80.00%.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

Warning

Review ran into problems

🔥 Problems

Git: Failed to clone repository. Please run the @coderabbitai full review command to re-trigger a full review. If the issue persists, set path_filters to include or exclude specific files.

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[Wirasm]

We are already running 0.2.121