fix(workflow-executor): align @forestadmin/* deps with the workspace

[hercemer42]

Summary

@forestadmin/workflow-executor:build is currently failing in CI (and on main) with a wall of TS2305/TS2339/TS2614 errors:

'@forestadmin/ai-proxy' has no exported member 'AiClient' / 'BaseChatModel' / 'BaseMessage' / 'DynamicStructuredTool' / …
'@forestadmin/forestadmin-client' has no exported member 'ServerUtils'
Property 'getOne' does not exist on type 'Collection'   (@forestadmin/agent-client)

Root cause: workflow-executor pins its @forestadmin/* deps to stale exact versions that lag the workspace and don't export the APIs its source uses. Every other package in the monorepo references the current workspace versions and builds fine — workflow-executor is the lone laggard. (It's a chicken-and-egg: semantic-release bumps these on release, but the broken build blocks the release that would fix them.)

Change

Bump the three pinned deps to the current workspace versions:

dep	from	to
@forestadmin/agent-client	1.5.10	1.6.0
@forestadmin/ai-proxy	1.10.1	1.11.0
@forestadmin/forestadmin-client	1.39.9	1.40.0

(plus the corresponding yarn.lock update).

Also: exclude claude-fable-5 from tool-supporting models (ai-proxy)

Fixing the build let the full test matrix run for the first time in a while, which unmasked a pre-existing failure in the required LLM Integration Tests (ai-proxy) job: Anthropic's new claude-fable-5 fails the tool-support sweep. It has always-on thinking — it rejects the thinking: {type: "disabled"} param that LangChain sends by default, and it requires thinking blocks to be replayed verbatim on multi-turn tool calls, which the proxy's OpenAI-format message conversion drops. Adding it to ANTHROPIC_UNSUPPORTED_MODELS gives users a clear AIModelNotSupportedError upfront instead of provider 400s (same pattern as the o1/o3/o4 exclusions on the OpenAI side). Proper Fable support would need thinking-block passthrough in the wire format — a separate feature.

This has to land here rather than in its own PR because the integration job can't go green on any branch until this PR's build fix is on it.

Verification

lerna run build --scope @forestadmin/workflow-executor --include-dependencies now succeeds (tsc clean) where it previously failed. ai-proxy unit tests (404) and lint pass; the previously failing datasource-mongo and OpenAI-503 failures cleared on re-run (flaky/transient).

This is independent of #1641 (PRD-496) — it unblocks the workflow-executor build on main for everyone.

🤖 Generated with Claude Code

Note

Align @forestadmin/* dependencies in workflow-executor with workspace versions

Bumps three direct dependencies in package.json to match the rest of the workspace: @forestadmin/agent-client to 1.6.0, @forestadmin/ai-proxy to 1.11.0, and @forestadmin/forestadmin-client to 1.40.0. The lockfile is updated accordingly.

Changes since #1643 opened

• Added 'claude-fable-5' to the ANTHROPIC_UNSUPPORTED_MODELS constant in the ai-proxy package with documentation explaining the incompatibility is due to always-on thinking requirements and mandatory verbatim replay of thinking blocks across tool calls, and added test cases verifying that isModelSupportingTools correctly identifies 'claude-fable-5' with provider 'anthropic' as unsupported while confirming 'claude-opus-4-8' remains supported [655833a]
• Reformatted comment documentation for the ANTHROPIC_UNSUPPORTED_MODELS constant [2e56cd2]

^{Macroscope summarized 13493e0.}

[qltysh]

Coverage Impact

This PR will not change total coverage.

🚦 See full report on Qlty Cloud »

🛟 Help

• Diff Coverage: Coverage for added or modified lines of code (excludes deleted files). Learn more.

• Total Coverage: Coverage for the whole repository, calculated as the sum of all File Coverage. Learn more.

• File Coverage: Covered Lines divided by Covered Lines plus Missed Lines. (Excludes non-executable lines including blank lines and comments.)

  • Indirect Changes: Changes to File Coverage for files that were not modified in this PR. Learn more.