test(uipath-agents): coded HITL / process / RAG / tracing tests (PR 4/5)

[radugheo]

Summary

Adds four coded-agent tests for the platform capabilities the existing suite never exercised — human-in-the-loop, process invocation, RAG via Context Grounding, and @traced redaction.

Task	Tier	What it covers
skill-agent-coded-hitl-create-task	e2e	LangGraph agent that escalates large expenses through interrupt(CreateTask(app_name="ExpenseReview", app_folder_path="Finance", ...)), compiled with a MemorySaver checkpointer so the interrupt can pause/resume.
skill-agent-coded-process-invoke	e2e	LangGraph agent that delegates work to an existing RPA process via interrupt(InvokeProcess(name="DataScraper", process_folder_path="Workflows", ...)).
skill-agent-coded-rag-langgraph	e2e	LangGraph RAG agent that retrieves passages from the company_docs Context Grounding index (ContextGroundingRetriever(index_name=..., folder_path="Shared")) and composes the answer with UiPathChat.
skill-agent-coded-tracing-redaction	e2e	Simple Function agent whose main is decorated with @traced(name=..., input_processor=..., output_processor=...) to redact passwords and tokens before they reach the trace store.

What each check script asserts

• HITL — imports of interrupt and CreateTask; interrupt(CreateTask(app_name="ExpenseReview", app_folder_path="Finance", ...)) call shape; .compile(checkpointer=MemorySaver()) (or InMemorySaver); bindings.json declares the app resource with ActivityName=create_async and DisplayLabel=ExpenseReview. Negative case verified: removing the checkpointer makes the check fail.
• Process invoke — interrupt(InvokeProcess(name="DataScraper", process_folder_path="Workflows", ...)) call shape; bindings.json process resource with ActivityName=invoke_async.
• RAG — ContextGroundingRetriever imported from uipath_langchain.* and constructed with index_name="company_docs" / folder_path="Shared"; bindings.json index resource with ActivityName=retrieve_async; both retriever and UiPathChat are lazy (no module-level construction).
• Tracing — main carries exactly one @traced(...) decorator with all three kwargs: name, input_processor, output_processor. Rejects the conflicting hide_input=True / hide_output=True on the same call site (the skill recommends processors over hide_* for partial visibility). Negative case verified.

Stacked on #474.

Test plan

• All four check_*.py dry-run green against synthetic well-formed projects.
• Negative cases verified: HITL without a checkpointer fails; tracing with hide_input=True alongside processors fails.
• All twelve task YAMLs in coded/ parse and tag lists are consistent.

[github-actions]

Claude finished @radugheo's task in 2m 34s —— View job

---

PR Review: test(uipath-agents): coded HITL / process / RAG / tracing tests

• Gather context and read review guidelines
• Read full diff and all changed files
• Check test task structure and conventions
• Check for secrets, cross-skill references, and link validity
• Post review

---

Summary

Adds four new coded-agent e2e test tasks covering human-in-the-loop (interrupt(CreateTask)), process invocation (interrupt(InvokeProcess)), RAG via Context Grounding (ContextGroundingRetriever), and @traced redaction. Each task has a YAML definition and a Python check script. The tests are well-structured, follow established patterns from existing coded tests, and make good use of shared utilities (_shared/bindings_assertions, _shared/ast_lazy_init_check).

Change-by-Change Review

1. tests/tasks/uipath-agents/coded/hitl_create_task/check_hitl_create_task.py

Severity: Low

Solid check script. Validates all four HITL primitives (interrupt import, CreateTask import + call shape, checkpointer, app binding). Good use of shared helpers.

One issue: unused import json on line 24. The script imports json but never uses it directly — JSON parsing is handled by load_bindings in the shared module.

Fix this →

2. tests/tasks/uipath-agents/coded/hitl_create_task/hitl_create_task.yaml

Severity: OK

Well-structured task YAML. Tags are consistent with repo conventions ([uipath-agents, e2e, coded, lifecycle:generate, feature:hitl-coded]). This is the only new task that includes a command_executed criterion for uip codedagent new (scaffold step) — the other three only check uip codedagent init. This is fine since the prompt says "scaffold → init", but worth noting the inconsistency (see #5 below).

Weight distribution is reasonable: scaffold (1.0) + init (1.5) + shape check (6.0).

3. tests/tasks/uipath-agents/coded/process_invoke/check_process_invoke.py

Severity: Low

Clean and focused. Validates interrupt(InvokeProcess(...)) call shape, parameter values, binding, and lazy-init.

Minor observation: the regex name\s*=\s*["\']DataScraper["\'] on line 54 would technically match a suffix like other_name="DataScraper". In practice the false-positive risk is negligible since the value DataScraper is unique to this context, but a \b anchor before name (i.e. \bname\s*=) would make it precise. Same pattern used in HITL for app_name and app_folder_path where the kwarg names are longer and inherently more specific.

4. tests/tasks/uipath-agents/coded/process_invoke/process_invoke.yaml

Severity: OK

Clean. Tags consistent. Only checks uip codedagent init (no scaffold step) — matches the other three non-HITL tasks.

5. tests/tasks/uipath-agents/coded/rag_langgraph/check_rag_langgraph.py

Severity: OK

Good two-stage import validation: first checks the class name is referenced at all, then verifies the import source is from uipath_langchain.*. The flexible import regex (uipath_langchain\.[\w.]+) correctly handles both uipath_langchain.agent.tools and uipath_langchain.retrievers.

6. tests/tasks/uipath-agents/coded/rag_langgraph/rag_langgraph.yaml

Severity: OK

Clean. Well-written prompt that specifies two-node architecture (retrieve + respond) with lazy-init requirement.

7. tests/tasks/uipath-agents/coded/tracing_redaction/check_tracing_redaction.py

Severity: Low

Good AST-based validation — properly parses decorator kwargs rather than relying on regex. The negative check for hide_input=True / hide_output=True alongside processors is a nice touch that enforces the skill's recommendation.

One observation on line 49: the import check uses string containment ("from uipath.tracing import traced" not in text and "uipath.tracing" not in text). This is looser than the regex-based import checks in the other scripts. A comment string containing uipath.tracing would pass the gate. Since the AST check downstream validates the actual @traced(...) decorator, this is not a functional issue — the import check is more of a diagnostic hint. But for consistency with the other check scripts, a regex like r"from\s+uipath\.tracing\s+import\s+[^\n]*\btraced\b" would be tighter.

8. tests/tasks/uipath-agents/coded/tracing_redaction/tracing_redaction.yaml

Severity: OK

Clean. The prompt correctly specifies a Simple Function agent (not LangGraph), matching the @traced decorator use case. No secrets — "password" and "token" references are instructional only.

What's Missing

• Scaffold (uip codedagent new) criterion inconsistency: hitl_create_task.yaml includes a command_executed check for uip codedagent new but the other three tasks (process_invoke, rag_langgraph, tracing_redaction) do not, despite all four prompts saying "scaffold → init". This is not a bug — the shape check scripts validate the project directory exists — but it's a coverage gap. Consider adding the scaffold criterion to the other three for consistency, or removing it from HITL. (Low)
• No smoke-tier task in this batch: All four tasks are tagged e2e. The repo already has a smoke task (simple_echo), so this isn't a gap per se, but worth noting that these features have no fast-feedback tier.

Area Ratings

Area	Status	Notes
Frontmatter	N/A	No skill SKILL.md changes
E2E Tests	✅ OK	Four well-structured e2e tasks with shape checks
Skill Body	N/A	No skill content changes
References & Assets	N/A	No reference changes
Repo Hygiene	✅ OK	No secrets, no cross-skill refs, CODEOWNERS covers parent path

Issues for Manual Review

• Verify that the bindings.json resource shapes asserted in the check scripts (e.g., app with ActivityName=create_async, process with ActivityName=invoke_async, index with ActivityName=retrieve_async) match the current UiPath SDK binding spec. The check scripts encode specific field names and values that could drift if the SDK changes.
• Confirm ContextGroundingRetriever is still importable from uipath_langchain.agent.tools and/or uipath_langchain.retrievers — the RAG check accepts both paths.

Conclusion

Clean PR. The four test tasks are well-structured, consistent with existing coded tests, and make good use of shared utilities. The check scripts are thorough — especially the tracing script's AST-based decorator validation and the HITL script's checkpointer assertion.

Three Low items worth addressing:

1. Remove unused import json in check_hitl_create_task.py:24
2. Consider \bname anchor in check_process_invoke.py:54 for regex precision
3. Consider tightening the string-containment import check in check_tracing_redaction.py:49 to a regex for consistency

None are blockers. Approve.

---