fix(examples): fix 9 broken examples + worker-tool registry lookup (companion to #438)

[ling-senpeng13]

Summary

Pending decision on whether to take these fixes.** Companion to #438 (verification report, docs only); together they replace #437 (closed). Every fix below reproduces on main and was verified end-to-end against a live conductor-oss 3.32.0-rc.8 server. Agents suite: 91 PASS / 12 ERROR / 3 HUNG before → 96 PASS / 2 ERROR / 2 HUNG after; tests/unit/ai: 1744 passed.

Counts at a glance

22 examples failed or misbehaved during verification (15 agents suite, 7 core suite). Disposition after this PR:

Outcome	Agents	Core	Total
✅ Newly passing — fixed by code changes here	4	5	9
⏱ Passes with a longer suite timeout (no code change needed)	1	0	1
➖ Reclassified — env-dependent → SKIP, or daemon/interactive by design	6	2	8
❌ Still failing — not fixed here (table at the bottom)	4	0	4
Total	15	7	22

(kitchen_sink.py is counted in "still failing": this PR fixes its TypeError but it then fails server-side.)

Example fixes

File	Bug	Fix
helloworld/greetings_worker.py	duplicate def greet shadowed the registered worker → spawn-pickling failure on macOS/Windows (broke workers_e2e, helloworld, task_listener_example)	rename second function greet_sync
worker_example.py	hardcoded /Users/viren/ metrics dir → PermissionError everywhere else	tempfile.gettempdir()
user_example/user_workers.py	imported examples.user_example.models while loaded as user_example.user_workers — spawn children crashed	consistent package-relative import
agents/kitchen_sink.py	@agent classifiers declared a required prompt arg; decorator invokes them with zero args → TypeError	zero-arg signatures
agents/16i, agents/16j	tools defined inside factory functions → SpawnSafetyError (<locals> not importable by spawn workers)	hoisted to module level
agents/94_openai_runner_tools.py	@function_tool rebinds the module global to an unpicklable FunctionTool	plain function at module level, function_tool() applied at Agent(...) construction
agentic_workflows/llm_chat.py	(a) two LLM tasks sent system-only messages — server fails the workflow before the first task; (b) example printed "Conversation complete." and exited 0 over a FAILED workflow	added user messages; exit non-zero on non-COMPLETED terminal state
agents/run_all_examples.py	6 env-dependent examples counted as ERROR	classified SKIP with reasons (dg skill repo; workflow messages API absent in rc.8)

SDK fix

_try_worker_task in conductor/ai/agents/tool.py: once ToolRegistry.register_tool_workers re-registers a @worker_task tool's task name with a spawn-safe ToolWorkerEntry wrapper, the raw identity check failed and a second get_tool_def() on the same tool raised TypeError (broke 14_existing_workers.py). The lookup now walks __wrapped__/fn_direct/fn_ref carriers. Two regression tests added.

Not fixed here — 4 examples (see the report in #438)

Of the 22 failures this verification surfaced, 9 examples are fixed by code changes in this PR, 9 are resolved without code fixes (reclassified or timeout), and 4 remain unfixed — the agents suite still shows 2 ERROR + 2 HUNG after these fixes:

#	Example	Status after this PR	Why it is not fixed here
1	agents/kitchen_sink.py	ERROR	This PR fixes its classifier TypeError, but it then fails server-side: HTTP 500 — Task translation_swarm defined as a sub-workflow has no workflow definition available. Sub-workflow (SWARM strategy) registration on conductor-oss rc.8 needs SDK/server investigation.
2	agents/74_cli_error_output.py	FLAKY (ERROR in suite run)	LLM-behavior assertion — the agent paraphrases stderr instead of quoting it verbatim, so it passes or fails depending on model output. Needs a looser assertion or stricter prompt, which changes the example's intent.
3	agents/86_coding_agent.py	HUNG	Does not finish within 420s even running solo. Root cause unknown; needs investigation (or reclassification as long-running).
4	agents/68_context_condensation.py	HUNG in suite / PASS solo	Passes solo in ~346s but exceeds the 360s suite timeout under 8-way contention. A code fix is arguably wrong — the suite timeout or example size is the issue.

[codecov]

Codecov Report

❌ Patch coverage is 87.50000% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
src/conductor/ai/agents/tool.py	87.50%	2 Missing ⚠️

Files with missing lines	Coverage Δ
src/conductor/ai/agents/tool.py	91.54% <87.50%> (-0.40%)	⬇️

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.