fix(hitl): align HITL skill and tests to v1.0 node schema

[dushyant-uipath]

Summary

• output not result: Updates all skill docs and eval tests to use $vars.<nodeId>.output (v1.0 renamed the outputId from result to output)
• cancelled/timeout handles removed: Cleaned up all references to non-existent handles; v1.0 exposes only completed
• fieldId vs variable rule documented: Added Critical Rule 9 to SKILL.md and a callout in hitl-node-quickform.md — output object keys are the field id, not field.variable
• 4 new developer-mistake quality tests guarding the most common HITL authoring errors:
  • quality_08: agent uses field ID (not variable name) for output access
  • quality_09: agent infers correct types (boolean/number/date) instead of defaulting to text
  • quality_10: binding on input fields only — output fields must have no binding
  • quality_11: inOut fields are pre-filled via binding AND appear in $vars.<nodeId>.output after submit
• TEST_PLAN.md rewritten with v1.0 invariants table, variable binding accuracy section, real-world consequence coverage map

Files changed

All files are under skills/uipath-human-in-the-loop/ and tests/tasks/uipath-human-in-the-loop/ (Dushyant-owned).

Test plan

• quality_08 — ContractReview: field ID vs variable name access
• quality_09 — ExpenseReview: boolean/number/date type inference
• quality_10 — SupplierOnboarding: binding direction correctness
• quality_11 — RCAReview: inOut pre-fill + downstream output access
• Existing quality_04, quality_07, e2e_01, e2e_03 pass with updated .output paths

[github-actions]

Claude finished @dushyant-uipath's task in 2m 58s —— View job

---

Summary

Aligns the HITL skill docs and eval tests to the v1.0 node schema: renames result → output, removes cancelled/timeout handles, documents the fieldId vs variable distinction (Critical Rule 9), and adds 4 new quality tests guarding common developer mistakes. Also updates cross-file links to the new shared/ paths in uipath-maestro-flow.

Change-by-Change Review

1. skills/uipath-human-in-the-loop/SKILL.md

Severity: OK
Two clean changes: adds Critical Rule 9 (fieldId not variable for output access) and updates the reporting template from .result → .output. Rule 9 is well-written and addresses a real agent mistake pattern.

2. skills/uipath-human-in-the-loop/references/hitl-node-quickform.md

Severity: OK
Comprehensive v1.0 alignment — all result → output changes are correct. Cross-file links updated to new shared/ paths (cli-commands.md, variables-and-expressions.md) — verified both targets exist. Definition entry updated: version: "1.0", shape: "square", cancelled/timeout handles removed, input handle constraints removed, outputDefinition keys corrected. The fieldId callout and runtime variables table are accurate and well-placed. The binding rule row correctly generalizes HITL outputId to output for v1.0.

3. skills/uipath-human-in-the-loop/references/hitl-node-apptask.md

Severity: OK
Edge wiring correctly updated to completed-only with clear "no cancelled or timeout handles in v1.0" note. Variables table and status description updated. Clean.

4. skills/uipath-human-in-the-loop/references/hitl-node-coded-action-app.md

Severity: High ❌
result → output and status description were correctly updated, but line 323 still reads:

wire completed, cancelled, timeout

This contradicts v1.0 (only completed available) and conflicts with the identical section updated in hitl-node-apptask.md:348. An agent following this reference will attempt to wire non-existent handles.

Fix: Change line 323 from:

**Edge wiring** — wire `completed`, `cancelled`, `timeout`. See [hitl-node-quickform.md](...) for edge format.

to:

**Edge wiring** — wire `completed` (only handle available in v1.0). See [hitl-node-quickform.md](...) for edge format.

Fix this →

5. tests/tasks/uipath-human-in-the-loop/TEST_PLAN.md

Severity: OK
Thorough rewrite: v1.0 invariants table, variable binding accuracy section, coverage map with real-world consequences, gap priority updates. Correctly removes the quality_08_timeout_handle gap (timeout handle no longer exists in v1.0). Status updated from Draft to Active.

6. tests/tasks/uipath-human-in-the-loop/e2e_01_invoice_approval_greenfield.yaml

Severity: OK
Clean .result → .output update in success criteria.

7. tests/tasks/uipath-human-in-the-loop/e2e_03_gdpr_compliance_greenfield.yaml

Severity: OK
Correctly updated: description no longer mentions timeout handle wiring, handles_wired hardcoded to ["completed"], file_contains check changed from "timeout" to "completed". The timeout_configured field (ISO 8601 duration on node input) is correctly preserved — it's the timeout duration property, not the handle.

8. tests/tasks/uipath-human-in-the-loop/quality_04_all_handles.yaml

Severity: OK
All .result → .output changes correct and consistent.

9. tests/tasks/uipath-human-in-the-loop/quality_07_runtime_vars.yaml

Severity: OK
All .result → .output changes correct and consistent.

10. tests/tasks/uipath-human-in-the-loop/quality_08_variable_binding_fieldid.yaml (new)

Severity: OK
Well-designed test: forces a field ID vs variable name distinction by giving id: "approved" / variable: "legalApproval" and validating the agent accesses .output.approved not .output.legalApproval. Includes both positive (file_contains with .output.approved) and negative (excludes with .output.legalApproval) checks. task_id, tags, and structure all follow conventions.

11. tests/tasks/uipath-human-in-the-loop/quality_09_dev_mistake_wrong_type.yaml (new)

Severity: Medium ⚠️
The test description claims it tests "boolean for yes/no decisions, number for amounts, and date for deadlines." The prompt includes a submittedDate input field described as "should be a DATE" and the report.json template asks for "submittedDate": "<type used>" — but there is no assertion in success_criteria that validates field_types.submittedDate equals "date". The claimed date-type coverage is not actually enforced.

Fix: Add an assertion to the json_check block:

      - expression: "field_types.submittedDate"
        operator: equals
        expected: "date"

Fix this →

12. tests/tasks/uipath-human-in-the-loop/quality_10_dev_mistake_binding_direction.yaml (new)

Severity: OK
Thorough test covering binding direction correctness. Good use of both file_contains (positive binding patterns) and json_check (report validation). max_turns: 90 is reasonable given the complex flow structure.

13. tests/tasks/uipath-human-in-the-loop/quality_11_inout_field_access.yaml (new)

Severity: OK
Well-structured test for inOut direction semantics. Correctly validates that inOut fields have binding (pre-fill) AND appear in output. The mock data ("Network timeout caused by misconfigured firewall rule") is realistic and domain-appropriate.

What's Missing

• hitl-node-coded-action-app.md:323 — cancelled/timeout handle references not cleaned up (see finding feature: Improve coded agent development from claude code experience #4 above)
• quality_09 date assertion — description promises date type coverage but doesn't enforce it (see finding UiPath development skill: orch, solutions, is #11 above)
• Nothing else identified. All relative links resolve, CODEOWNERS is in place, no secrets or cross-skill references.

Area Ratings

Area	Status	Notes
Frontmatter	OK	No changes to frontmatter, existing frontmatter valid
E2E Tests	OK	4 new quality tests well-structured; existing tests correctly updated
Skill Body	OK	Critical Rule 9 is well-placed and accurate
References & Assets	Issue	coded-action-app.md:323 still references removed handles
Repo Hygiene	OK	CODEOWNERS present, no secrets, changes scoped to HITL skill

Issues for Manual Review

• v1.0 schema accuracy: The outputDefinition in the definition entry was changed to "source": "=result.Action" for status. Verify this matches the actual HITL runtime's output contract — an incorrect source expression would cause runtime variable resolution failures.
• typeVersion on node vs version on definition: The node JSON example uses "typeVersion": "1.0.0" (line 82 of quickform.md) while the definition uses "version": "1.0". Confirm these are intentionally different fields with different version granularity.

Conclusion

Solid, well-scoped PR that correctly aligns the HITL skill to v1.0. The new quality tests are thoughtfully designed and cover real agent failure patterns. Two issues to fix before merge:

1. High: Update hitl-node-coded-action-app.md:323 to remove cancelled/timeout handle references — this is the last inconsistent spot in the v1.0 migration
2. Medium: Add the missing submittedDate date-type assertion to quality_09 to match the test's claimed coverage

---

• Read review criteria and project conventions
• Analyze full diff (origin/main...HEAD)
• Read existing PR conversation
• Review each changed file in full context
• Check skill-specific requirements (links, cross-refs, secrets, CODEOWNERS)
• Post review findings

[github-actions]

Claude finished @dushyant-uipath's task in 2m 34s —— View job

---

Review: PR #529 — fix(hitl): align HITL skill and tests to v1.0 node schema

• Read review criteria and project conventions
• Read full diff (origin/main...HEAD)
• Read existing PR conversation
• Review each changed file in full context
• Check skill-specific requirements (links, cross-refs, secrets, CODEOWNERS)
• Post review findings

---

Summary

Aligns the HITL skill docs and eval tests to the v1.0 node schema: renames result → output, removes cancelled/timeout handles, documents the fieldId vs variable distinction (Critical Rule 9), and adds 4 new quality tests guarding common developer mistakes.

Change-by-Change Review

1. skills/uipath-human-in-the-loop/SKILL.md

Severity: OK
Adds Critical Rule 9 (fieldId not variable for output access) and updates the reporting template from .result → .output. Rule 9 is well-written — addresses a real agent mistake pattern with a clear explanation of the variable property's actual role.

2. skills/uipath-human-in-the-loop/references/hitl-node-quickform.md

Severity: OK
Comprehensive v1.0 alignment:

• All result → output changes correct (variables.nodes entries, outputDefinition, binding rule row, runtime variables table, code example)
• Definition entry updated: version: "1.0", shape: "square", input handle constraints removed, cancelled/timeout handles removed from handleConfiguration
• outputDefinition keys corrected: "result" → "output" with "var": "output", status source changed to "=result.Action"
• Cross-file links updated to new shared/ paths — verified both cli-commands.md and variables-and-expressions.md exist at the target paths
• fieldId callout and runtime variables table accurately document the v1.0 access pattern

3. skills/uipath-human-in-the-loop/references/hitl-node-apptask.md

Severity: OK
Edge wiring correctly updated to completed-only with explicit "no cancelled or timeout handles in v1.0" note. Removed the multi-edge JSON example showing cancelled/timeout edges. Runtime variables table and variables.nodes description both updated from result to output.

4. skills/uipath-human-in-the-loop/references/hitl-node-coded-action-app.md

Severity: OK
All four changes are correct: edge wiring updated to completed only, variables.nodes description uses output, runtime variables table uses output and updated status description. The prior automated review flagged line 323 — it's fixed in this commit.

5. tests/tasks/uipath-human-in-the-loop/TEST_PLAN.md

Severity: OK
Thorough rewrite with v1.0 invariants table, variable binding accuracy section, coverage map with real-world consequences, and gap priority updates. Status updated from Draft to Active. Correctly removes the quality_08_timeout_handle gap (handle no longer exists). File inventory accurately reflects all present and missing test files.

6. tests/tasks/uipath-human-in-the-loop/e2e_01_invoice_approval_greenfield.yaml

Severity: OK
Clean .result → .output update in the file_contains success criterion (line 96).

7. tests/tasks/uipath-human-in-the-loop/e2e_03_gdpr_compliance_greenfield.yaml

Severity: OK
Correctly updated: handles_wired hardcoded to ["completed"], file_contains check changed from "timeout" to "completed". The timeout_configured field (ISO 8601 duration on node input) is correctly preserved — it's the timeout duration property, not the handle.

8. tests/tasks/uipath-human-in-the-loop/quality_04_all_handles.yaml

Severity: OK
All .result → .output changes correct and consistent.

9. tests/tasks/uipath-human-in-the-loop/quality_07_runtime_vars.yaml

Severity: OK
All .result → .output changes correct and consistent.

10. tests/tasks/uipath-human-in-the-loop/quality_08_variable_binding_fieldid.yaml (new)

Severity: OK
Well-designed test: forces a field ID vs variable name distinction by giving id: "approved" / variable: "legalApproval" and validating the agent accesses .output.approved not .output.legalApproval. Includes both positive (file_contains with .output.approved) and negative (grep -L legalApproval) checks. Task ID, tags, and structure all follow conventions.

11. tests/tasks/uipath-human-in-the-loop/quality_09_dev_mistake_wrong_type.yaml (new)

Severity: OK
Tests boolean, number, and date type inference. The prior automated review flagged a missing submittedDate assertion — it's present in this commit (line 88-90). All four type assertions are covered (approved → boolean, amount → number, approved_amount → number, submittedDate → date).

12. tests/tasks/uipath-human-in-the-loop/quality_10_dev_mistake_binding_direction.yaml (new)

Severity: OK
Thorough test covering binding direction correctness. Good use of both file_contains (positive binding patterns) and json_check (report validation). Validates that input fields have binding and that approved_has_binding is false. max_turns: 90 is reasonable given the complex flow structure with decision routing.

13. tests/tasks/uipath-human-in-the-loop/quality_11_inout_field_access.yaml (new)

Severity: OK
Well-structured test for inOut direction semantics. Correctly validates that inOut fields have binding (pre-fill) AND appear in output. The mock data is realistic and domain-appropriate. Tests both the direction: "inOut" declaration and the downstream access pattern.

What's Missing

Nothing identified. All relative links resolve, CODEOWNERS is in place, no secrets or cross-skill references beyond the expected uipath-maestro-flow dependency. The two issues from the prior automated review (coded-action-app.md:323 handles and quality_09 date assertion) are both fixed in this commit.

Area Ratings

Area	Status	Notes
Frontmatter	OK	No changes to frontmatter, existing frontmatter valid
E2E Tests	OK	4 new quality tests well-structured; existing tests correctly updated
Skill Body	OK	Critical Rule 9 well-placed and accurate
References & Assets	OK	All three reference files consistently updated for v1.0
Repo Hygiene	OK	CODEOWNERS present, no secrets, changes scoped to HITL skill

Issues for Manual Review

• v1.0 schema accuracy: The outputDefinition in the definition entry uses "source": "=result" for output and "source": "=result.Action" for status. Verify these match the actual HITL runtime's output contract — an incorrect source expression would cause runtime variable resolution failures.
• typeVersion vs version granularity: The node JSON example uses "typeVersion": "1.0.0" (line 82 of quickform.md) while the definition uses "version": "1.0". Confirm these are intentionally different fields with different version granularity — typeVersion on the node instance vs version on the definition entry.

Conclusion

Clean, well-scoped PR. All result → output renames are consistent across skill docs and tests. The cancelled/timeout handle cleanup is complete — no stale references remain. Critical Rule 9 (fieldId vs variable) is a valuable addition that addresses a real agent failure pattern. The 4 new quality tests are thoughtfully designed with both positive and negative assertions. The two issues flagged by the prior automated review are already fixed in this commit. No issues found — this is ready to merge.