feat(uipath-maestro-flow): add child simulation docs and skill test

[AAgnihotry]

Summary

• Add --parent flag docs to the evaluate capability for simulating individual tools inside agent nodes
• Child simulations let you mock individual tools (e.g. Web_Search, Send_Email) inside an agent node instead of the whole agent
• A single command is enough — the parent simulation is auto-created if it does not exist
• Companion PR for CLI: UiPath/cli#3923

Changes

• CAPABILITY.md: add child simulation to "When to use" and "Common tasks"
• commands-reference.md: add --parent to simulation add/list/remove flag tables, examples, and output codes (FlowEvalChildSimulationAdd/List/Remove)
• eval-sets-guide.md: add "Child Simulations (Agent Tool Simulation)" section

Skill test

• child_simulation_crud.yaml: full lifecycle — add Static + Llm child, list, remove — with auto-created parent, 11 success criteria
• check_child_simulation_crud.py: verifier reads eval-set JSON and checks parent auto-creation, child persistence, and removal

Test plan

• Skill test passes: child_simulation_crud task runs end-to-end
• Existing simulation_crud task still passes (no breaking changes to docs)

Ran skill-flow-eval-child-simulation-crud locally and it passed.

🤖 Generated with Claude Code

[github-actions]

Claude finished @AAgnihotry's task in 3m 43s —— View job

---

PR Review: feat(uipath-maestro-flow): add child simulation docs and skill test

• Read review criteria and project conventions
• Analyze full diff
• Read all changed files in full context
• Check PR conversation for resolved issues
• Apply skill-specific checks (links, cross-refs, secrets, test structure)
• Post review findings

---

Summary

This PR adds documentation for the --parent flag on uip maestro flow eval simulation add/list/remove, enabling child tool simulations inside agent nodes. It touches three reference docs (CAPABILITY.md, commands-reference.md, eval-sets-guide.md) and adds a smoke test task with a Python verifier script. The docs are well-structured, the test is thorough, and no issues block merge.

Change-by-Change Review

1. skills/uipath-maestro-flow/references/evaluate/CAPABILITY.md

Severity: OK

Two clean additions: one bullet in "When to use" (line 15) and one row in the "Common tasks" table (line 99). Both link to the correct anchors in eval-sets-guide.md#child-simulations-agent-tool-simulation and commands-reference.md#simulations. Verified the anchor targets exist.

2. skills/uipath-maestro-flow/references/evaluate/commands-reference.md

Severity: OK

Well-executed changes:

• --component-type correctly updated from required to optional with a clarifying note about --parent defaulting it to Node (line 144).
• --output-schema auto-resolution note correctly scoped to "top-level simulations only; required when using --parent" (line 148).
• New --parent flag row is well-documented with auto-creation behavior (line 149).
• Child simulation example is clear with the "no separate parent step" comment (lines 189–203).
• simulation list and simulation remove correctly show --parent in both the flag table and examples (lines 213, 237).
• Output Codes table updated with FlowEvalChildSimulationAdd/List/Remove (line 336).

3. skills/uipath-maestro-flow/references/evaluate/eval-sets-guide.md

Severity: Low

New "Child Simulations (Agent Tool Simulation)" section (lines 220–265) is well-structured. Covers add (Static + Llm), list, remove, auto-creation semantics, and --component-type defaulting.

One minor note: the section repeats the --output-schema example for Send_Email that's already shown in commands-reference.md (lines 237–243 here vs. lines 159–163 in commands-reference.md). Per Reference Pointer Discipline, this is borderline — but since eval-sets-guide.md is a guide (not a pointer doc), having a self-contained example is acceptable. Not flagging as an issue.

4. tests/tasks/uipath-maestro-flow/evaluate/child_simulation/child_simulation_crud.yaml

Severity: OK

Solid test design:

• Tags follow convention: [uipath-maestro-flow, smoke, mode:build, lifecycle:generate, feature:eval, feature:child-simulation] — skill name first, valid tier.
• task_id: skill-flow-eval-child-simulation-crud follows the skill-<domain>-<capability> pattern.
• 11 success criteria with appropriate weights (1.0 scaffolding → 3.0 primary child sim commands → 3.0 JSON verification).
• Good use of command_not_executed with negative lookahead to verify no standalone parent simulation was added (line 136).
• run_command criterion at line 140 delegates deep verification to the Python script — the right pattern.
• Prompt is goal-oriented but includes steps (slightly more hand-holding than ideal for a smoke test, but this is testing a brand-new --parent flag so the specificity is justified).
• Inherits agent: from experiment config — no redundant overrides. Clean.

feature:child-simulation is a new tag leaf name — appropriate to propose in this PR per test-writing rules.

5. tests/tasks/uipath-maestro-flow/evaluate/child_simulation/check_child_simulation_crud.py

Severity: OK

Thorough verifier that checks:

1. Eval-set JSON exists with name == "Agent Tools"
2. Data point search-test present in evaluations[]
3. Parent simulation agent-1 exists with componentType == "agent" and simulationStrategy == "Llm" (auto-creation defaults)
4. childSimulations array exists on the parent
5. Web_Search child is present (add worked)
6. Send_Email child is absent (remove worked)

Good defensive coding: handles missing dirs, missing JSON, alternate key names (componentId / component_id / id). Error messages are actionable.

What's Missing

• No passing-run claim in the PR body. The test-writing rules (.claude/rules/test-writing.md step 5) require running the task with coder-eval and adding a passing-run claim to the PR. The PR body's test plan has unchecked boxes. This is common for new tasks and doesn't block merge, but should be resolved before or shortly after.

Area Ratings

Area	Status	Notes
Frontmatter	OK	Not modified — existing frontmatter unchanged
E2E Tests	OK	Smoke test added; this PR extends an existing skill, not a new one
Skill Body	OK	Three reference docs updated consistently
References & Assets	OK	All links verified, anchors resolve, no orphaned files
Repo Hygiene	OK	No secrets, no cross-skill refs, changes scoped to one skill

Issues for Manual Review

• CLI flag accuracy: The --parent flag behavior (auto-creating parent with type agent and strategy Llm, --component-type defaulting to Node, --output-schema not auto-resolving for child sims) is documented per the companion CLI PR (UiPath/cli#3923). Reviewers with CLI access should verify these defaults match the actual implementation.
• Output code names: FlowEvalChildSimulationAdd/List/Remove should be confirmed against the CLI's actual JSON output.

Conclusion

Clean PR. Documentation is well-structured, consistent with existing simulation docs, and follows repo conventions. The test task is thorough with a good Python verifier. No blocking issues found — approve once the passing-run claim is added.

---

[github-actions]

Claude finished @AAgnihotry's task in 2m 58s —— View job

---

Coder-eval task lint (advisory)

1 task YAML changed; verdicts: 0 Critical, 1 High, 0 Medium, 1 Low, 0 OK.

Rubric: .claude/commands/lint-task.md. This check is advisory and never blocks merge.

Evidence of passing run

❌ High — PR body does not claim the changed/added tasks have been run and passed. The "Test plan" section contains unchecked boxes (- [ ] Skill test passes: child_simulation_crud task runs end-to-end), which describe intent, not a completed run. Please edit the PR description to add a line like:

Ran skill-flow-eval-child-simulation-crud locally and it passed.

Per-task lint

tests/tasks/uipath-maestro-flow/evaluate/child_simulation/child_simulation_crud.yaml — verdict: Low

Issues:

• [Low] Prompt over-specification: lines 58–59 explicitly state "Do NOT add a separate parent simulation … The --parent flag auto-creates the parent." This is the key behavior the skill should teach — leaking it in the prompt means an agent could pass without consulting the skill's child-simulation docs. The corresponding command_not_executed criterion (lines 133–138) then verifies compliance with an instruction the prompt handed over, not skill knowledge. Consider trimming to "add child simulations on parent agent-1 — use the child simulation workflow from the skill" and letting the skill docs teach auto-creation.

Suggested fixes:

• Remove or soften the explicit auto-creation hint on line 59. Replace with a goal statement (e.g., "add child simulations targeting agent-1 without a separate parent step") so the agent must learn the --parent auto-creation from the skill.

Within-PR duplicates

No duplicate clusters detected. The new child_simulation_crud.yaml tests child-level simulation CRUD (--parent flag, childSimulations nesting, parent auto-creation) while the existing sibling simulation/simulation_crud.yaml tests parent-level simulation CRUD (--component-type, top-level simulations array). Materially distinct operations sharing a scaffold — good template reuse.

Conclusion

⚠ 1 task has issues, max severity High (evidence of passing run). The per-task lint is Low only. Advisory only — not blocking merge.

---