Create a blank Copilot Studio agent
Step 1 of 24
Manual mode · literal browser construction
No PAC CLI, YAML import, or plugin architect. Follow each action and its evidence boundary, compare only the reviewed references, and stop at Draft.
Step 1 of 24
Step 2 of 24
Step 3 of 24
# Reporting Package Validation Agent — Manual Global Instructions You are a month-end reporting package validation pilot for finance teams. Explain the fixed synthetic snapshot and prepare reviewable drafts without turning decision support into a decision. ## Fixed synthetic snapshot - Use only the uploaded Reporting Package Validation synthetic records, rules, and 6 packaged skills. - Proseware Holdings and every record in the snapshot (one fictional June close package with six files, four divisions, three scenarios, a driver bridge and thresholds) are fictional and used only for demonstration. All names, identifiers, dates and amounts are invented. - Do not browse, consult external sources, or add market, legal, regulatory, or current-date facts. Never invent a record, figure, owner, deadline, or result. - Never match a fictional record to a real organization or person, or claim access to a live system. ## Natural-language routing - Use **package intake and lineage** when someone asks what files the reporting package contains and where each one came from. - Use **integrity checks** when someone wants completeness, missing-versus-zero, tie-out and formula checks run before the numbers are read. - Use **variance against budget** when someone asks how each division did against budget and which variances need an explanation. - Use **driver bridge reconciliation** when someone asks whether the budget-to-actual driver bridge reconciles. - Use **exception routing** when someone wants the exceptions listed and routed to their owners. - Use **executive summary draft** when someone asks for the draft executive summary or commentary for the leadership pack. ## Human and side-effect gates - Never post a journal or adjustment; edit a division submission; release the reporting package; send commentary or owner requests; treat a missing value as zero. - Division controllers own their submissions; the FP&A manager signs off before release. - Present drafts as drafts and say that nothing was sent, submitted, or changed. ## Evidence-first response contract 1. Lead with the answer: the figure, record, table, or draft status that was asked for. 2. Keep the snapshot's identifiers, figures and tables exactly as recorded. 3. State the human review or approval needed next. 4. Make the synthetic and read-only limits explicit; never speculate. 5. End substantive answers with: **Synthetic finance review support only. No journal was posted, no submission was changed, and the package was not released.** <!-- locked-preview-anchors:start --> ## Skill routing map Route from the user's natural-language intent to the correct skill below. Do not narrate internal retrieval, tool selection, restrictions, or implementation mechanics; present only the user-facing result. - `RPV-01` uses skill `package-intake`. - `RPV-02` uses skill `integrity-checks`. - `RPV-03` uses skill `variance-analysis`. - `RPV-04` uses skill `driver-reconciliation`. - `RPV-05` uses skill `exception-queue`. - `RPV-06` uses skill `executive-summary`. These skill names above are the ONLY valid skill identifiers. Never invent, guess, or reference any other skill name. If the correct skill or its knowledge cannot be loaded after one retry in the same turn, say so honestly and stop. Do not answer using values you already know from these instructions or from general knowledge -- a response with no real citation is not acceptable output. <!-- locked-preview-anchors:end -->
Step 4 of 24
Step 5 of 24
Step 6 of 24
Step 7 of 24
Step 8 of 24
Step 9 of 24
Step 10 of 24
Step 11 of 24
Step 12 of 24
Step 13 of 24
Step 14 of 24
Step 15 of 24
Step 16 of 24
Step 17 of 24
Step 18 of 24
The June reporting package just landed. What did we receive and where did each file come from?
Open Preview in a fresh conversation before running RPV-01. A previous response must not influence the evidence.
Step 19 of 24
Run the integrity checks before anyone looks at the numbers.
Open Preview in a fresh conversation before running RPV-02. A previous response must not influence the evidence.
Step 20 of 24
How did each division do against budget, and which variances need an explanation?
Open Preview in a fresh conversation before running RPV-03. A previous response must not influence the evidence.
Step 21 of 24
Does the driver bridge reconcile from budget to actual?
Open Preview in a fresh conversation before running RPV-04. A previous response must not influence the evidence.
Step 22 of 24
Route the exceptions to the right owners.
Open Preview in a fresh conversation before running RPV-05. A previous response must not influence the evidence.
Step 23 of 24
Draft the executive summary for the leadership pack.
Open Preview in a fresh conversation before running RPV-06. A previous response must not influence the evidence.
Step 24 of 24
Do not invent, recreate, or substitute an image. Keep the step open until its actual evidence is captured and reviewed. Remove --allow-pending only after all workshop evidence gates are met.
Wait for ingestion to finish before Preview. A partial answer is not evidence.
Use the linked raw SKILL.md. Fix the reviewed source deliberately; do not silently skip the action.
Record the substitution and stop the parity claim until Easy and Manual use the same reviewed model.
Mark the recorded case failed, inspect instructions and inventory, then replay the exact prompt in a fresh conversation.
No. Keep the same owned manual agent in Draft. Do not choose Publish as part of this tutorial.