uipath-planner

UiPath Planner — Design & Task Derivation

Two jobs, one entry point:

1. Design — turn a Process Definition Document (PDD) into an implementation-ready Solution Design Document (SDD). Select scope (single product or multi-project Solution), write the SDD.

2. Plan — derive the per-skill task list from an SDD (or a non-PDD request), route to specialists, emit live TaskCreate calls.

Never execute the work. Outputs are SDD markdown, plan/tasks markdown, and live tasks — implementation always routes to a specialist.

The skill has three paths, decided by the Entry Guard:

• Phase D — Design. Input is a PDD, or an explicit "design / architect this" request. Author the SDD; the SDD write ends the turn, and Lane A continues on the next turn. See sdd-generation-guide.md. Case Management designs run through Phase D's Case Design Lane (case-design-lane-guide.md) — a conversational design ending in ONE Case Review confirmation.
• Lane A — PDD-driven. Input is an SDD with the ## Planner Handoff marker (written by Phase D, or hand-written). Read it, derive tasks, emit live tasks. Zero to two user prompts. See pdd-driven-lane-guide.md.
• Lane B — Non-PDD. No SDD; a non-PDD multi-project request. Elicit preferences, detect project type, write a plan, emit live tasks. 0–3 prompts (5-call cap). See non-pdd-lane-guide.md.

When to Use This Skill

• User provides a PDD or any process-knowledge source — a PDD, Confluence page, BPMN model, meeting/Zoom transcript, SOP, or requirements doc (as PDF / Word / Markdown / .txt / .bpmn / pasted) — and asks to design or build from it → Phase D
• User asks to design / architect / generate an SDD for a UiPath automation → Phase D
• User provides an SDD path → Lane A
• The request is non-trivial — spans separate buildable projects that each need their own specialist (a Flow orchestrating standalone RPA processes or agents that must themselves be built) → Lane B
• The request is ambiguous — no single specialist clearly matches, or "what can I build?"

Skip this skill for single-project tasks — load the specialist directly. A request is single-project (one specialist owns it end-to-end) even when it bundles several things inside one project: a Flow with script nodes plus an inline HITL approval step plus its own solution wrapper is one uipath-maestro-flow task. Inline nodes (HITL QuickForm, script, connector, inline agent) and solution scaffolding are author sub-steps the specialist performs itself — not separate skills to orchestrate. Counting them as distinct skills and emitting a plan is the most common mis-trigger. This skill is only for work spanning separate buildable projects (distinct .uipx projects), or for turning a PDD into an architecture.

Critical Rules

1. Plan & design only — never author automation code. Outputs: SDD markdown (Phase D), plan/tasks markdown (Lanes A/B), and live TaskCreate calls. NEVER write XAML, C#, Python, JSON, or project/scaffold files. Implementation always routes to a specialist. (SDD/plan authoring is the only file authoring this skill does.)

2. Run the Entry Guard first. Inspect the input and route to Phase D / Lane A / Lane B before anything else.
3. Select scope before designing architecture (Phase D) — and gate every product by platform constraints. Single product (RPA Process/Library/Test Auto, Maestro Flow, Maestro BPMN, Case, Agents, Coded Apps, API Workflows) vs multi-project Solution determines the template(s) and project structure. Use the Product Selection Guide: Constraint Gate → Level 1 → 1.5 (RPA sub-type) → 1.75 (Solution composition) → 2.5 (project decomposition). The delivery model (Cloud / Automation Suite / standalone — asked at Phase D entry) blocks unavailable products via platform-availability-guide.md; user-excluded products are never re-offered.

4. The SDD is architecture only — no task lists. Phase D produces the SDD (Project Structure, Data Definitions, Testing Strategy, …). Task derivation is Lane A's job. Never put Task 1 / Task N templates or implementation TaskCreate calls in the SDD. End the SDD with a ## Next Steps section. (Progress-tracking TaskCreate calls are a separate, allowed use.)

5. Write the ## Planner Handoff header AND the <!-- planner-handoff:v1 --> marker into every SDD. Load-bearing detection contract — the Entry Guard detects either signal (redundant on purpose). Generated by: uipath-planner. Fields: Status (draft → ready), Execution autonomy, Delivery model, SDD scope, the solution-root block (solution scope only: Solution root SDD, Solution ID, Project SDD role root|child, Independently executable: no on children — the root's Project Inventory + SDD Index are the sole executable entry; children and root share ONE canonical tasks file), Project list section, Tasks file, Generated by, Generation date, Template validation. The first incremental write stamps Status: draft; only after SME resolution AND the template-superset check does Phase D flip it to ready (+ Template validation: passed) — the marker identifies a planner SDD, the Status field says whether it is consumable. Lane A derives tasks from ready only (missing field = legacy → treat as ready). Every template places the header after ## Document History.
6. Honour the template section structure as a hard superset contract. Write single-product scope to <PROCESS_NAME_KEBAB>-sdd.md; write Solution scope to a <SOLUTION_NAME_KEBAB>-solution-sdd.md overview plus one <PROJECT_NAME_KEBAB>-sdd.md per project. If the user specifies an output path for the SDD, use it instead of these defaults. After writing, diff the generated H2/H3 headings against the template TOC — the generated set MUST be a superset. A missing template-required H2/H3 is an SDD defect, not an [SME REVIEW] item — regenerate it.
7. Testing is mandatory and thorough — never offer "happy path only". Phase D writes a full Testing Strategy section (happy path, edge cases, error scenarios, e2e for Master Projects). The plan adds a mandatory Testing task per generation skill, routing to that specialist's testing references — never describing the procedure inline. Implementation specialists may scope down at execution time; the SDD and plan do not.
8. Route — do not redescribe. The plan says WHICH skill to load and IN WHAT ORDER. It does NOT describe specialist-internal flows (target configuration, OR registration, XAML pipelines, HITL field/outcome schema, auth, testing procedures). For a HITL step, pass business intent only ("manager approves or rejects an expense; can add a reason if rejected") — never a field-level spec; the HITL specialist chooses the schema shape.
9. Per-phase prompt budget. Phase D runs under its own checkpoint model (see sdd-generation-guide.md) — no hard numeric cap. Lanes A and B each cap at 5 AskUserQuestion calls. Ask execution autonomy exactly once (Phase D entry) and write it into the handoff header; Lane A reads it and never re-asks. Scope/UI answers resolved in Phase D flow forward via the SDD.
10. Fill gaps with [DEFAULT] or [SME REVIEW] — never silently invent business rules. [DEFAULT] for industry-standard patterns (retry counts, timeouts); [SME REVIEW] for business-knowledge gaps. Resolve [SME REVIEW] items with the user before finalizing (they gate the Status: ready flip, not the Phase 3 Step 0 skeleton write). For Agent/Coded App gaps, use AskUserQuestion (proceed-with-gap-filling vs different product) — never auto-fallback.
    • These two tokens are the entire gap vocabulary, and their exact string is load-bearing. Every downstream consumer — the Action Required — SME Review Items block, the gap counts, the Status: ready flip, Lane A's assumption carry-over, and any reviewer grepping the file — finds these items by exact match on [SME REVIEW] / [DEFAULT].
    • Write the marker verbatim, with nothing else inside the brackets. The explanatory note goes after the closing bracket: | UiPath version | [SME REVIEW] not recorded in the PDD |. Never fold the note inside ([SME REVIEW — not recorded]) and never substitute another word ([PLACEHOLDER], [TBD], [TODO], <fill in>, N/A, prose like "to be confirmed"). A decorated or renamed marker is invisible to every consumer above — the gap silently disappears.
    • When you map or copy content out of another document (a source SDD into a client template, a PDD into an SDD, one SDD section into another), carry its markers through byte-for-byte. Re-wording a marker you did not author destroys a gap someone else recorded. Add new markers for the gaps you find; never edit the ones already there.

11. The terminal artefact of a Solution build is a packed .uipx. The SDD's ## Next Steps section points the user at the uipath-solution skill (uip solution init → project add per project → resources refresh → pack). A bare project folder is not the deliverable. Exception: when the Constraint Gate blocks Solutions for the delivery model — standalone, Automation Suite older than 2.2510, or a user exclusion — rewrite Next Steps to per-package Orchestrator publish routed via uipath-platform.

12. Never copy SDD architecture into the plan, and never invent selectors or UI targets. The plan references SDD section paths in skill prompts; it does not duplicate architecture content. Selectors require application inspection at development time — leave them for the specialist.
13. Sole author of Case Management SDDs. Every case design path — conversational, build handoff from uipath-maestro-case, draft finalization, PDD-driven — runs the Case Design Lane (case-design-lane-guide.md), which owns the conversation, the one confirmation, and the write; the case model and its defaults live in case-design-layers-guide.md and the render contract in the case SDD template. Design time resolves identities only — never schema discovery, project scaffolding, or create-on-missing; those are build concerns, recorded as gate decisions. Open the lane's files with the Read tool, never cat. The design reading set is exactly three files — the lane guide, the layers guide, and the case SDD template. case-sdd-conformance-checklist.md is NOT one of them: it belongs to the gate subagent at the end. A design-only request (stops before caseplan.json) seeds sdd.md to disk as the FIRST tool call after the reading set, before any design reasoning — the seed's content depends on none of it — then appends section by section. The lane's § Sketch bounded no-build item 0 owns the cadence.

Entry Guard

Run this guard before anything else.

0. Product design-lane signals — checked first. Some products register a
   dedicated Phase D design lane (today: Case Management —
   references/case/case-design-lane-guide.md). Route to Phase D — Design through
   the product's lane when any of these fire (the path taxonomy stays the three
   paths above; a lane is Phase D's product-specific entrance):
   - a build request handed off by that product's build skill because its
     design input is missing (same conversation) → the lane's Build
     handoff mode,
   - a design request for that product with no PDD → the lane, direct,
   - the product's design draft exists (or is named) and the user asks to
     finalize it → the lane, draft finalization.

1. No document path?
   - Explicit design/architect language ("design this", "architect this",
     "generate an SDD"), OR an inline-described process with enough detail to
     substitute for a PDD (process steps + applications + exceptions) → Phase D — Design.
     A design/SDD request that is thin on process detail still routes to Phase D —
     elicit the missing steps/applications/exceptions via `AskUserQuestion`
     (Phase 1) rather than degrading to a Lane B plan.
   - Otherwise → Lane B — Non-PDD. Lane B is the default for document-less
     multi-project requests.

2. Document path → read its first ~50 lines.
   - Contains `## Planner Handoff` OR `<!-- planner-handoff:v1 -->` → Lane A — PDD-driven.
     (Either signal alone is sufficient — redundant on purpose.)
   - Reads as a PDD **or any process-knowledge source** — a process
     description (steps + applications + exceptions) in a PDD, Confluence
     page, BPMN model, meeting/Zoom transcript, SOP, or requirements doc;
     or a binary .pdf/.docx the user calls a PDD → Phase D — Design. Less-structured
     sources (transcripts, thin wikis) still route to Phase D but trigger
     heavier Phase 1 elicitation.

3. Otherwise (no marker, ambiguous, or unparseable) — ask via AskUserQuestion:

   > What is the document at <path>?
   > 1. Process Definition Document (PDD) — author the SDD (Phase D — Design), then derive tasks
   > 2. Solution Design Document (SDD) — proceed with task generation (Lane A — PDD-driven)
   > 3. Other context — read it; use its content to resolve Lane B elicitation
   >    questions (skip any question it answers) and as plan input (Lane B — Non-PDD)

4. Route per the choice. For an SDD with no handoff header, proceed with safe
   defaults — see pdd-driven-lane-guide.md Step 1 for the default set and how
   defaults are surfaced to the user.

Do not pattern-match on filename or extension alone; those are unreliable. The ## Planner Handoff heading and the <!-- planner-handoff:v1 --> marker are the load-bearing detection contract — Phase D writes both deterministically; the guard detects either.

Phase D — Design (summary)

When triggered: input is a PDD, or an explicit design/architect request. Three phases; full detail in sdd-generation-guide.md. All user questions use numbered-choice format.

1. Phase 1 — PDD Analysis & Scope Selection. Ask execution mode (Autonomous or Interactive) and delivery model (Cloud / Automation Suite / standalone) in one batched call — skip the delivery question when the PDD or request states it, or when the uip login status preflight resolves it from the session BaseUrl. Read the full PDD, extract structured information (including environment & constraint signals, as-is/to-be), synthesize the need profile (Step 3.5 — product picks reason from the need, not keywords), then run Constraint Gate → Level 1 (primary scope) → Level 1.5 (RPA sub-type) → Level 1.75 (Solution composition) → Level 2.5 (project decomposition). Step 2.5 runs an authenticated uip library search (CLI auth required). In Interactive mode, present a summary with the recommended scope at the top and alternatives below; in Autonomous mode, proceed.
2. Phase 2 — Architecture Review. Load the product-specific template. Generate the architectural core sections. Present for review in Interactive mode.
3. Phase 3 — Full SDD Generation. Step 0 is a hard gate — write the SDD skeleton to disk (header + ## Planner Handoff header/marker with Status: draft + ## Decisions Made / ## Recommended Scope + the Phase 1/2 sections) BEFORE generating any Phase 3 section, then append each remaining section as you generate it, so a watchdog kill still leaves a gradeable file. The disk write is progress task 4, never the last task. Generate all remaining sections including the thorough Testing Strategy. Resolve [SME REVIEW] items before finalizing (unresolved items carry recorded defaults and do not block ready; only architecture-blocking items keep draft). The LAST write flips Status to ready after the superset check passes. The SDD write is a turn boundary: in autonomous mode, end the turn after the SDD summary and continue into Lane A on the next turn (do not stack both phases in one turn).

Lane A — PDD-driven (summary)

When triggered: an SDD with the Planner Handoff marker is detected (or Phase D just wrote one).

1. Read the SDD's ## Planner Handoff header. Status: draft → refuse task derivation (unfinished Phase D or a blocking SME item — offer resume or regenerate; missing field = legacy → ready; open default-carried SME items on a ready SDD travel into tasks as assumptions). Reuse the execution autonomy chosen in Phase D — do not re-ask. If SDD scope: solution, run the root algorithm: resolve the Solution root, verify every indexed child (exists, same Solution ID, ready), read every child's architecture, merge shared resources, emit ONE canonical tasks file (see pdd-driven-lane-guide Step 3).
2. If <process>-tasks.md already exists, ask continue / regenerate (1 prompt). See plan-and-tasks-format.md → Regenerate logic.
3. Parse the SDD project list section. Pick the multi-skill pattern.
4. Ask the UI batch (3 questions, 1 call) only if the SDD's Application Inventory lists UI applications and the answers aren't already resolved.
5. Derive tasks. Write <process>-tasks.md.

6. If Execution autonomy: interactive → EnterPlanMode for review. If autonomous → emit live tasks directly.
7. Emit TaskCreate calls + addBlockedBy edges. Hand off.

Full procedure: pdd-driven-lane-guide.md.

Lane B — Non-PDD (summary)

When triggered: no SDD; a document-less multi-project request (the default route when no explicit design/architect language or inline-described process points to Phase D).

1. Step 1 — detect before asking (no prompts): read any provided context doc, filesystem detection (Glob/Read/Grep — cross-platform), multi-skill pattern classification, need-driven project-type inference, delivery-model resolution.
2. Step 2 — single-skill exit: one project owned end-to-end by one specialist → stop Lane B, load that specialist directly with the detected context (no plan file, no tasks).
3. Step 3 — batched elicitation: generation approach + execution autonomy + project-type fallback (when still vague) + delivery model (when a gated product is a candidate) in one AskUserQuestion call. Drop any question already resolved.
4. Step 4 UI batch — only when the plan includes UI automation in uipath-rpa.

5. Write YYYY-MM-DD-<feature>.md to docs/plans/ (project) or ./plans/ (no project). Every task prompt embeds the plan path.
6. If explore-first → EnterPlanMode. If simultaneous → emit plan as text + live tasks.

Full procedure: non-pdd-lane-guide.md.

Skill capability map

High-level view of what each specialist owns. Do not describe internal flows of any specialist in your plan — each skill documents its own procedures and will drift out of sync if duplicated here.

Skill	What it owns	Handles auth?	Handles deploy?
uipath-rpa	RPA workflows (XAML and C# coded): create, edit, build, run, debug. Owns all UI automation authoring end-to-end, including live-app exploration and probing.	No (relies on Studio)	No — defer to uipath-solution for .uipx multi-project, uipath-platform for single non-solution packages
uipath-agents	AI agents — code-based (LangGraph / LlamaIndex / OpenAI Agents) and low-code (agent.json)	Yes (uip login)	Yes — end-to-end
uipath-coded-apps	Web apps (.uipath/ dir): build, sync, package, publish, deploy	Yes (uip login)	Yes — end-to-end
uipath-functions	Coded Functions — Python (uip function new -l py; Pydantic typed I/O) or TypeScript/JavaScript (-l ts / js, TypeScript default; defineFunction + schema-first typed I/O) — atomic deterministic logic: transforms, custom-auth API calls, ERP/IS-connection queries. Runs serverless or on an RPA robot; a JS/TS function is delivered as a run-as-job function or an HTTP endpoint (Coded App backends) — one calling mode per function. Invoked from Maestro (Flow/BPMN), agents-as-tools, Orchestrator API, or its HTTP trigger. No LLM/agent loop (that → uipath-agents)	Yes (uip login)	Yes — uip function pack/publish
uipath-rules	Business Rules projects — one .dmn decision table (DMN 1.5) per project: uip rules init, refresh, validate, debug. Called by BPMN business rule tasks, Case business-rule tasks, and RPA	Yes (uip login for debug)	No — uip solution pack/publish/deploy via uipath-solution
uipath-maestro-flow	.flow files orchestrating RPA, agents, apps	Yes (uip login)	Partial — Orchestrator deploy of .uipx-wrapped solutions → uipath-solution; non-solution single package → uipath-platform
uipath-maestro-bpmn	.bpmn files — standards-based BPMN 2.0 process orchestration (gateways, events, boundary timeouts, subprocesses, multi-instance) over RPA, agents, APIs, HITL. Authors its own inline userTask/HITL nodes.	Yes (uip login)	Yes — packages, uploads, publishes/deploys, and runs via uip maestro bpmn; Orchestrator deploy of .uipx-wrapped solutions → uipath-solution
uipath-maestro-case	Case Management authoring — caseplan.case from an SDD (case pack compiles it to caseplan.case.bpmn; authoring .bpmn directly is uipath-maestro-bpmn)	Yes (uip login)	Partial — publishes to Studio Web and Orchestrator itself; .uipx solution deploy → uipath-solution
uipath-api-workflow	API Workflows (JSON document.dsl): author, run locally (uip api-workflow run), connector activities	Yes (uip login)	No — uip solution pack/publish via uipath-solution
uipath-human-in-the-loop	HITL node authoring — approval gates, escalations, write-back validation inside Flow projects. Coded-agent HITL → uipath-agents; BPMN / Case / RPA own theirs inline (userTask / task type / Action Center)	No (authoring only)	No — ships inside the host project
uipath-platform	Auth (uip login), Orchestrator (folders, processes, jobs, machines, users, roles), resources (assets, queues, storage buckets + bucket files, libraries, webhooks, triggers), Integration Service (connectors, connections, activities, IS triggers), Data Fabric entities/records/files/choice-sets (uip df), traces, licensing	Yes (auth hub)	Yes — for non-solution single packages and Orchestrator-side post-deploy ops
uipath-ixp	Document Understanding / IXP — extraction from semi-structured documents (invoices, forms): taxonomy, model, classify / extract / validate. Standalone project, or the extraction step a primary consumes	Yes (uip login)	Partial — model publish/tag via uip ixp; consumed by the host project
uipath-connector-builder	Integration Service custom connector authoring (REST+JSON) on disk via uip is connectors builder — build a connector when the catalog has none for a required integration (needed by API Workflows / Maestro / Agents; RPA can call the API directly instead)	Yes (uip login)	Yes — import/publish is deployment (uip is connectors)
uipath-mcp-servers	UiPath AgentHub MCP server registration (6 types: uipath, coded, command, remote, swagger, platform) and resource-tool authoring on uipath-type servers (automation, agent, agentic-process, api-workflow). Wraps Orchestrator resources, external HTTP MCP endpoints, OpenAPI specs, published coded agents, local subprocess commands, or first-party UiPath services as MCP tools. NOT for FastMCP / Python mcp SDK work.	Yes (uip login)	Yes — registration is deployment (posts directly to AgentHub)
uipath-solution	uip solution lifecycle (init, pack, publish, deploy, activate) for .uipx solutions. Runs as the final skill in PDD-driven flows (deploy of .uipx solutions).	Yes (uip login)	Yes — for multi-project Solution (.uipx) deploys

Reference Navigation

Phase D — Design

File	Purpose
SDD Generation Guide	Phase orchestrator — Phase 1, 2, 3 step-by-step instructions
Case Design Lane Guide	The case design conversation — entry, tenant grounding, authoring policy, the one Case Review, and the write
Case Design Layers Guide	The case model and every design Default — skeleton (triggers, stages, the ten task types), gates (lifecycle rules, exits, sequencing, secondary lanes), data (variables, outputs, expressions), time (SLAs, escalations, responses), naming rules, and the closure checklist
PDD Analysis Guide	How to extract structured data from PDDs in any format
Business Rules Guide	Validation vs decision vs reasoning rules; DMN-style decision tables; reasoning list. Load when filling RPA §4 or BPMN §5.
Product Selection Guide	Constraint Gate (delivery model + exclusions filter), Level 1 (primary scope), Level 1.75 (Solution composition), Level 2.5 Part B (cross-product project list merge), Level 3 (capability add-ons), template mapping
Platform Availability Guide	Product × delivery-model availability matrix (Cloud / Automation Suite / standalone), alternatives for blocked products, verification rule for uncertain cells. Load whenever the Constraint Gate runs — every delivery model: "Cloud is not uniform", variants (GovCloud / Dedicated / Test Cloud) and per-tenant entitlements can block products inside the Cloud column.
RPA Product Guide	RPA Level 1.5 (sub-type), Level 2 (authoring mode), Level 2.5 Part A (RPA decomposition), R-07 naming, REFramework. Load when Level 1 = RPA or a Solution includes RPA.
Package Selection Guide	NuGet package selection; Integration Service vs NuGet rules; per-product dependency manager. Load when filling §14 Packages or equivalent.
Tenant Library Search Guide	Step 2.5 procedure for discovering deployed libraries via uip or libraries list + JMESPath filtering — auth preflight, ranking, zero-results branch, manual fallback.
Attended Re-auth Pattern Guide	Design contract for portal automations with a human-only login (hardware 2FA token, smart card, biometric): handoff point, state-verified resume, attended robot, routing to uipath-rpa. Load when the PDD signals a non-scriptable interactive sign-in.
ASDD Crosswalk Guide	Generate the client-facing Word SDD/ASDD: warn the user for the template path, match the markdown SDD into its sections, compute missing pieces. Includes the crosswalk for the standard UiPath SDD template. Load at Step 2.5 when the user wants the official document.

SDD templates

File	Purpose
RPA Template	SDD template for RPA Process / Library / Test Automation
Flow Template	SDD template for Maestro Flow
BPMN Template	SDD template for Maestro BPMN
Case Management Template	SDD template for Case Management (downstream Case Definition Blueprint shape)
Agent Template	SDD template for UiPath Agents
Coded App Template	SDD template for Coded Apps (web)
API Workflow Template	SDD template for API Workflows

Scripts (Phase D)

Script	Purpose
scripts/docx-extract.{sh,ps1}	.docx PDD → UTF-8 markdown + extracted screenshots (pandoc). Run at Phase D Step 1 when a Word PDD renders garbled — never drive Word via COM. Cross-platform twins: bash …/docx-extract.sh on macOS/Linux/Git-Bash; pwsh -File …/docx-extract.ps1 (or powershell -File … for Windows PowerShell 5.1) on Windows. Keep the two files behaviorally identical.
scripts/sdd-to-docx.{sh,ps1}	Generated SDD markdown → styled .docx, optional corporate --reference-doc. Run at Phase D Step 2.5 only when the user asks for Word output. Same cross-platform twin invocation (bash .sh / pwsh -File .ps1).

Lanes A & B — Planning

File	Purpose
PDD-driven Lane Guide	Lane A end-to-end — read SDD header, parse project list, derive tasks, write tasks.md, emit live tasks
Non-PDD Lane Guide	Lane B end-to-end — elicitation, project-type inference, filesystem detection, UI batch, write plan.md
Multi-skill Patterns Guide	The named multi-skill patterns (RPA build+deploy, Flow with local/deployed resources, Agent with RPA tools, etc.). Used by both lanes.

| Plan and Tasks Format | Header schema, task row schema, identity tuple, status states, regenerate-with-preservation algorithm, TaskCreate mapping, anti-hallucination rule, quality rules |

Anti-patterns

1. Skipping the entry guard. Always inspect the input first. A PDD silently treated as a generic doc produces a degraded deliverable.
2. Writing automation code or modifying the project. SDD and plan/tasks markdown only. In explore-first Lane B mode, non-mutating uip discovery is the upper limit.
3. Treating a single-project request as a plan — the most common mis-trigger. The inline-nodes rule lives in the Skip paragraph under When to Use This Skill.
4. Copying the PDD structure into the SDD. The SDD reorganizes content for implementation — it does not mirror the PDD's document flow.
5. Defaulting to RPA Process when the need profile points elsewhere. Use the Product Selection Guide decision tree on the synthesized need, not keywords: genuine judgment/reasoning → Agents; stages/SLA/approval lifecycle → Case Management ("AI" over a rule-expressible decision is still RPA). Forcing single-product scope when the PDD describes multiple coordinated projects is the same mistake — offer Solution.
6. Generating an Implementation Plan / task list inside the SDD. Architecture only; the SDD ends with ## Next Steps. Task derivation is Lane A's job.
7. Describing specialist-internal flows in the plan or SDD. Target configuration, OR registration, scaffolding, auth, pack/publish, testing procedures, HITL field schema — all owned by the specialist's own docs. Inlining creates drift.
8. Asking about test coverage depth. Testing is always thorough. The implementation specialist scopes down at execution time if the user wants a quick MVP; the planner does not offer the option.
9. Recommending a skill that contradicts filesystem signals. .flow → uipath-maestro-flow, not uipath-rpa.
10. Inventing selectors from screenshots, or asking the user what the planner / library / filesystem can already answer. Default first; spend a prompt only on decisions only the user can make.
11. Renaming the ## Planner Handoff heading or stripping the <!-- planner-handoff:v1 --> marker. Either signal alone is sufficient for detection, but both should remain — removing both breaks Lane A detection silently.
12. Recommending a product the customer's platform cannot run, or re-offering an excluded one. Defaulting to cloud-modern picks (Maestro, Coded Apps, Document Understanding) on an Automation Suite engagement is the most expensive design error — every blocked product survives review only to fail at delivery. Run the Constraint Gate; once a user says "we don't want X", X stays blocked for the whole session.