# Compile User Map > Compile an OEM Modbus PDF or structured register map plus measurement intent into an organized user map, JSON, CSV, and optional target outputs in one resumable run. Use when the user wants an organized user map or offline outputs from an OEM source rather than a specialist review chain. - Skill: `studioxvii/compile-user-map` (Agent Skill, multi-file: 6 files) - Install (CLI): `npx skillmds@latest add studioxvii/compile-user-map` - Raw SKILL.md: https://api.skillmd.com/api/skills/studioxvii/compile-user-map/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Data & Analytics - License: Apache-2.0 - Author: studioxvii (https://skillmd.com/u/studioxvii) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/studioxvii/compile-user-map --- # Compile User Map Produce the requested outcome without exposing internal stages. Follow `../../references/interaction-contract.md`. ## Process 1. Read `references/request.md` and the interaction contract together. Use supplied skill, source, and case paths directly; a full plugin or case-directory inventory is unnecessary for a supported request. 2. Select an available Python 3.11+ executable. PDF input also requires `pdfplumber`; if it is unavailable, report the missing dependency and stop instead of installing software during the workflow. Structured input needs only the standard library. 3. Put the OEM source, measurement intent, optional targets, and any resume input in one request file. For a curated intent, follow the bounded source-inspection and exact-match procedure in `references/request.md`; do not assume that prose alone selects points. 4. Run ` /scripts/run.py --request --output `. 5. Inspect the result plus the user-map holds and target statuses against the user's requested outcome. `offline-complete` is valid only when source coverage is complete, selected PDF fields have confirmed source evidence, and selected points have no blocking holds. Return the state, elapsed time, artifact paths, exclusions, and one useful next step; never treat `next_action: none` alone as proof of completion. If a new case awaits selection, the result and grouped selection packet are the relevant handoff evidence. Do not reread the full OEM map or run a second integrity inspection of the just-created case unless those artifacts expose a discrepancy. 6. When the result needs a decision, present its complete grouped packet once, encode the reply in a new request, and rerun the same case. Do not invent fields or bypass case, source, packet, or artifact hashes. On resume, run `scripts/inspect_case.py ` first. It verifies the checkpoint and every indexed artifact without changing files, and returns the current case hash and packet. Use the resume example in `references/request.md`; its selection preparation helper copies verified bindings while requiring the actual chosen IDs, then the normal wrapper applies the reply. Implementation-code inspection is unnecessary for a normal supported request. 7. Continue this skill automatically for safe internal stages. When a typed decision, physical read, or target choice is required, recommend continuing this skill with the exact case and input. Do not expose internal specialist-stage choreography. ## Output files Open these first: - `output/user-map.md` - The short human-readable map organized by measurement group. - `output/user-map.csv` - The spreadsheet-ready map for people and common tools. - `output/user-map.json` - The same map in the complete machine-readable format. Normally leave these alone: - `targets/` - Open this folder only when you requested files for Node-RED, Modpoll (BETA), or ModScan (BETA). It contains the files to import into that tool. - `compile-result.json` - the agent reads this receipt to tell whether the run finished or needs something from you. You normally do not need to open it. - `case.json` - This checkpoint lets the agent continue the same job later without starting over. Keep it until the job is finished, and do not edit it. - `artifacts/` and `control/` - These let the agent verify and resume the job. Keep them for troubleshooting; ignore them during normal use. Completion requires `compile-result.json` plus the complete offline user-map bundle. Requested targets may remain independently held without invalidating completed outputs. A `partial` result still contains useful map files. Report those files and the one correction needed. Ask for input only when the state is `awaiting-source-decision`, `awaiting-selection-decision`, `awaiting-binding`, `awaiting-physical-read`, or `awaiting-byte-order-decision`. A corrected source starts a new case; do not describe it as a resume. ## Stop - Never perform a live device read. - Stop for unsupported writes, broadcasts, scans, credentials, and polling. - Do not invent fields or bypass case, source, packet, or artifact hashes. - Do not expose internal specialist-stage choreography. ## Finish - The requested offline bundle is usable: say `Done`, name its human, JSON, and CSV artifacts, then offer target generation only as an optional goal. - The current case can advance safely: continue it without a handoff. - A decision or external action blocks the requested outcome: recommend continuing `compile-user-map` with the case-bound packet or probe and invite `proceed`. - A requested target is held while the offline map is usable: report both facts and recommend the one action that unlocks that target. Never perform a live device read. Return a case-bound probe when physical evidence is required. Keep unsupported writes, broadcasts, scans, credentials, and polling stopped.