🍒 [v1.201] feat(hitl-skill): add Case Management surface — hitl-casetask-action.md + SKILL.md routing

[github-actions]

What

Adds Human-in-the-Loop support for the Case Management surface to the uipath-human-in-the-loop skill. Supersedes #541 (closed — same branch, force-pushed after a rebase onto latest main plus fixes described below; GitHub won't allow reopening a closed PR whose branch was force-pushed).

Changes

SKILL.md

• Adds Case surface detection: if caseplan.json is present (root type case-management:root), route to hitl-casetask-action.md
• Step 3 now presents two options for Case: QuickForm and App-based
• Step 5 has a Case section linking to the reference doc

references/hitl-casetask-action.md (new file)
Two paths, confirm before writing:

Path	When to use
QuickForm	Structured form fields, no deployed app. Schema lives in a separate .hitl.json file alongside caseplan.json. Action task gets context[] entries: hitlType: "quick", _schemaFileId, hitlSchemaId.
App-based	Existing deployed Action Center app. hitlType: "custom" + =bindings.* references.

Key implementation detail for QuickForm: the .hitl.json file uses a unified fields[] array where each field has a direction property ("input" / "output" / "inOut").

Also documents that every action task written via direct JSON — QuickForm or App-based — needs entryConditions[] (e.g. current-stage-entered) on the task node itself. This is a general Case Management structural requirement, not HITL-specific, but its absence from the doc caused the agent to inconsistently add it, which surfaced as a uip maestro case validate failure (Task has no entry rules) during local verification below.

Eval tests

• smoke_09_case_quickform.yaml — QuickForm lifecycle: creates caseplan.json, adds a QuickForm task, checks .hitl.json file + context entries
• quality_12_case_path_selection.yaml — Path selection for a data-enrichment scenario: should pick QuickForm (not App-based)

Verification (actually run locally, not just described)

Ran both new tests plus a regression sample of 8 existing uipath-human-in-the-loop smoke/quality tests via coder-eval against this branch:

Task	Result
skill-hitl-smoke-case-quickform (new)	SUCCESS, 1.000
skill-hitl-quality-case-path-selection (new)	SUCCESS, 1.000
7 existing smoke/quality regression tests	SUCCESS, 1.000 each
skill-hitl-smoke-neg-admin	Pre-existing known-flaky llm_judge test (0.4), already slated for deletion via #1920/#1921 — unrelated to this change

Along the way, fixed two test-authoring bugs found in the original tests (not skill bugs): path: "*.hitl.json" isn't glob-matched by this repo's coder-eval version (treated as a literal filename), and a grep -L bindings-check had non-portable exit-code semantics between BSD and GNU grep. Both criteria now use find-based run_command checks matching this repo's existing convention (see e.g. uipath-coded-apps tests). Also added metadata.caseExitRules to both tests' seed fixtures — a pre-existing case-level completion-rule requirement unrelated to HITL that a bare-bones seed caseplan.json was missing.

Manually confirmed the actual produced artifacts match the documented shapes:

• .hitl.json has fields[] + direction (not separate inputs/outputs/inOuts arrays)
• caseplan.json action task has data.context[] with hitlType: "quick", _schemaFileId, hitlSchemaId
• No data.formType or inline data.schema in caseplan.json
• root.data.uipath.bindings[] unchanged for the QuickForm path

🤖 Generated with Claude Code

---

Automatically cherry-picked from #2056 to release/v1.201 by @dushyant-uipath.

[dushyant-uipath]

Superseded by #3108, which carries this same cherry-pick plus a follow-up fix (#3107). Closing to avoid confusion.