Stakeholder Workshop Prep

Good workshops are won before anyone enters the room. The facilitator who has digested the prior research, knows what each stakeholder cares about, and arrives with a timed plan and ready materials runs a different session from the one improvising off a blank whiteboard. The difference is not talent; it is preparation time, and preparation time is the thing most teams cannot find.

Most of that preparation is synthesis of material that already exists: research reports nobody reread, analytics dashboards nobody exported, strategy decks with the answer buried on slide 34, meeting notes where the real disagreement is visible if you look. Pulling it together, citing it, and shaping it into a pre-read, an agenda, and a set of templates is exactly the kind of patient assembly an agent workflow does well — and exactly the work that gets skipped when the workshop is on Thursday.

This workflow runs in stages: an evidence agent builds the pre-read pack with sources, a stakeholder agent maps who is in the room and what they care about, an agenda agent drafts the session plan, and a materials agent produces the boards, printouts, and parking-lot doc. The facilitator reviews every stage and owns the room. The agents prepare; they do not facilitate, and they do not psychoanalyze anyone.

Reach for it whenever a session matters enough to prepare for and the inputs already exist somewhere: project kickoffs, journey-mapping sessions, prioritization workshops, design sprint days, and discovery kickoffs with a new client. It works for a 90-minute remote session and for a full sprint day; only the agenda agent's brief changes.

It is a Foundation workflow because the stages run one after another and each produces a document a facilitator can read in minutes. No browser automation, no scripts — just careful reading, structured output, and your review.

• A kickoff or discovery session where the team has more background material than anyone has read.
• A journey-mapping workshop that needs evidence on the wall, not opinions.
• A prioritization session where the pre-read should settle the factual questions before the room opens.
• An agency engagement where the prep is also the first impression.

Each stage produces one artifact, and each artifact has a clear quality bar: the pre-read cites its sources, the stakeholder map only uses material the team provided, the agenda has timings and a plan B, and the materials are ready to import or print without further formatting.

The prep runs as a Claude Code dynamic workflow, but unlike a sweep it is staged rather than wide: the agenda depends on the evidence and the stakeholder map, and the materials depend on the agenda. When the prompt includes the word workflow, or you run with /effort ultracode, Claude writes a JavaScript orchestration script that runs in the background, holding the digested source material in script variables and passing each stage's output to the next without flooding Claude's context with every page of every report.

The capacity numbers — up to sixteen concurrent subagents and a thousand per run — matter less here than the structure; the evidence agent may fan out one reader per source document, but the stages themselves run in sequence. Runs are resumable, so adding one late document does not restart the whole prep. Once the output fits how you facilitate, save it to ~/.claude/workflows/ as a personal /workshop-prep command, or to the project's .claude/workflows/ if the whole team preps the same way; the subagent definitions live in .claude/agents/ next to it.

The staging is also where the facilitator's review fits. You read the pre-read and the stakeholder map before the agenda is worth reading, and you read the agenda before the materials are worth generating. Reviewing in stages takes less time than reviewing one large output at the end, and it catches wrong assumptions before they propagate.

The workflow is only as good as what you give it, and the discipline of gathering inputs is itself useful prep. Put everything in one folder: research reports, analytics exports, strategy docs, the org chart or attendee list, recent meeting notes, and a short brief from you stating the workshop's goal and the decision it needs to produce.

The brief is the most important file and the shortest. A workshop without a stated decision tends to produce a pleasant conversation and a photo of sticky notes; the agenda agent will keep anchoring on whatever you write here, so write it carefully.

workshop-prep/
└── claims-journey-mapping-2026-06/
    ├── brief.md                  # goal, the decision the session must produce, constraints, date, duration
    ├── attendees.md              # names, roles, departments — provided by you, not researched by the agent
    ├── inputs/
    │   ├── claims-research-2025.pdf
    │   ├── nps-verbatims-q1.csv
    │   ├── claims-funnel-analytics.csv
    │   ├── service-strategy-2026.pdf
    │   └── steering-meeting-notes/
    ├── output/
    │   ├── pre-read.md           # evidence agent
    │   ├── stakeholder-map.md    # stakeholder agent
    │   ├── agenda.md             # agenda agent
    │   ├── boards/               # materials agent: FigJam/Miro CSVs and markdown boards
    │   └── parking-lot.md
    └── review-notes.md           # what you corrected at each stage

The prompt sets the stages, names the inputs, and draws the line that matters most: the stakeholder map is built only from material the team provided. No searching the web for attendees, no inferred personalities, no speculation beyond what the org chart and the meeting notes actually say.

Run this as a workflow.

Prepare the journey-mapping workshop described in workshop-prep/claims-journey-mapping-2026-06/brief.md.

Stage 1 — Evidence: read everything in inputs/ and write output/pre-read.md: what we already know about the claims journey, organized by journey stage, with a source citation (file and page or row range) for every claim, plus an explicit list of open questions the workshop should answer. Maximum 4 pages.

Stage 2 — Stakeholders: from attendees.md and the steering meeting notes only, write output/stakeholder-map.md: who is in the room, what each person or department is accountable for, what they are likely to want from this session, and where the notes show disagreement or tension. Use only the provided material; if the material does not say, write "unknown". Do not infer personality, motives, or anything not evidenced in the inputs.

Pause for my review of stages 1 and 2 before continuing.

Stage 3 — Agenda: draft output/agenda.md for a 3-hour session with 14 attendees: timed blocks, exercises, who speaks when, facilitation notes, an energy check after lunch logistics, and a plan B if the mapping exercise overruns.

Stage 4 — Materials: produce output/boards/ as FigJam-ready CSVs (one row per sticky: text, section, color) for the journey map skeleton and the assumptions board, a printable one-page exercise instruction sheet, and output/parking-lot.md.

Do not contact anyone, do not look up attendees online, and do not present any of this as final — I review and own every artifact.

The evidence agent is the stage most worth defining as a reusable subagent, because its quality bar — every claim cited, open questions listed, no padding — is the same for every workshop you will ever prepare.

---
name: workshop-evidence-writer
description: Digests prior research, analytics, and strategy documents into a short pre-read pack with a source for every claim. Read-only over inputs.
tools: Read, Grep, Glob, Write
---

You write the pre-read pack for a workshop from the documents in the inputs folder.

Rules:
- Every factual claim carries a citation: file name plus page, slide, or row range.
- Distinguish findings (evidenced) from hypotheses (someone's stated belief in a document) and label them.
- Surface disagreements between sources instead of smoothing them over; they are usually the most useful content in the room.
- End with "Open questions for the session" — the things the inputs cannot answer.
- Keep it under 4 pages. If the inputs support more, cut by importance to the stated decision, not by recency.
- Do not invent data, do not generalize from a single verbatim, and do not include anything you cannot cite.

The facilitator's review between stages is short but real. On the pre-read: are the citations right, and are the open questions actually open? On the stakeholder map: does it match what you know about the room, and is anything in it something you would not want an attendee to read over your shoulder? On the agenda: would you actually run this, and where will you deviate?

Then send the pre-read to attendees a few days ahead, with the open questions framed as the session's work. Keep the stakeholder map private to the facilitation team — it is preparation for facilitation, not a document about people to circulate.

• Pre-read goes to all attendees ahead of the session; it is the shared factual floor.
• Stakeholder map stays with the facilitator; it informs the agenda, not the room.
• Agenda timings are a plan, not a contract; the facilitation notes should say what to drop first.
• Materials get a dry run: import the CSV into FigJam once before the day.

An insurance company's claims service team booked a three-hour journey-mapping workshop with 14 attendees across claims operations, contact center, digital product, and underwriting. The prep inputs were two research reports, a quarter of NPS verbatims, funnel analytics, and eight months of steering meeting notes; the staged run took a little over two hours, including the facilitator's reviews.

The evidence agent's pre-read condensed roughly 240 pages into four, organized by journey stage, with 47 cited claims and nine open questions. The stakeholder map, built only from the meeting notes and the attendee list, surfaced one tension the facilitator had missed: operations and digital product were each tracking time-to-settle with different definitions, and both numbers appeared in the strategy deck. The agenda agent placed a definitions exercise in the first 30 minutes because of it.

On the day, the definitions exercise consumed 25 minutes and saved the afternoon — the room could not have mapped the journey honestly while disagreeing about what its core metric meant. The facilitator's verdict afterward: the agents did not run the workshop, but the prep meant the three hours were spent mapping rather than discovering what everyone already knew.

A product group preparing a quarterly prioritization workshop used the workflow mainly for the evidence stage. The pre-read pulled together delivery status, usage analytics, and the original goals for eleven candidate initiatives, with citations.

Two of the eleven did not survive the pre-read. One initiative's stated user problem had been resolved by a platform change two quarters earlier — visible in the analytics export, never connected to the project. Another existed in three planning documents with three different owners and no activity for five months. Both were withdrawn by their sponsors after reading the pre-read, before the workshop, which freed almost an hour of agenda for the initiatives where the evidence was genuinely contested.

The team kept a rule from this: the pre-read circulates at least three working days ahead, because its value is what people do with it before the room, not in it.

A design agency preparing a discovery kickoff with a new ecommerce client received the usual scatter of inputs: a brand strategy deck, a previous agency's audit, analytics access, and notes from three sales calls. The stakeholder agent, working only from the call notes and the client's attendee list, flagged that marketing's stated success metric (revenue per session) and product's (activation of a new loyalty program) pointed in different directions, and that the deck cited both without reconciling them.

The agenda agent turned that into the kickoff's centerpiece: a success-metrics alignment exercise before any journey or feature discussion, with a board template the materials agent had already built. The conflict surfaced in the room within 20 minutes — politely, with the evidence on the wall rather than as an accusation — and the engagement's measurement plan was agreed in the same session instead of unraveling in month two.

The agency's facilitator made one correction during review that mattered: the stakeholder map had inferred slightly too much about one attendee's seniority from a job title, and the facilitator rewrote that line from their own knowledge. That is the review working as intended.

Weak prep output reads like a generic facilitation guide; strong prep output could only be about this workshop, this room, and this decision. The fastest quality check is to ask of every line: which input did this come from?

The workflow prepares the room; it does not read it. Facilitation — sensing energy, deciding when to push and when to park, handling the senior voice that dominates, changing the plan at the coffee break — is the facilitator's craft and stays that way. The agenda is a starting position, and the facilitation notes should make the planned deviations explicit.

The stakeholder map deserves its own boundary, stated twice because it matters: it is built only from material the team provides and it never speculates about people beyond that material. It maps accountabilities and documented positions, not personalities. If it ever reads like a psychological profile, delete it and tighten the agent's instructions before the next run.

• It cannot facilitate, read the room, or resolve conflict; it can only make conflict discussable with evidence.
• It cannot know what is missing from the inputs; the facilitator's domain knowledge fills the gaps at review.
• It must not profile attendees beyond provided material, and the map stays with the facilitation team.
• It cannot make the workshop's decision; it makes sure the decision is the thing the room spends its time on.

Save the staged run as /workshop-prep once the artifacts match how you facilitate, keep the subagent definitions in .claude/agents/, and reuse the folder structure for every session. The second prep takes half the time of the first, and the tenth one is mostly gathering inputs and reviewing — which is the part that was always worth a facilitator's attention.

1. Write the brief: the workshop's goal and the decision it must produce.
2. Gather the inputs into one folder: research, analytics, strategy docs, attendee list, meeting notes.
3. Run the evidence agent: a pre-read pack under 4 pages, every claim cited, open questions listed.
4. Run the stakeholder agent: accountabilities, stated positions, and tensions — from provided material only.
5. Review both; correct from your own knowledge and note what you changed.
6. Run the agenda agent: timed plan, exercises, facilitation notes, plan B.
7. Run the materials agent: FigJam/Miro CSV boards, printable instructions, parking-lot doc; dry-run the import.
8. Send the pre-read days ahead, keep the stakeholder map private, run the room yourself, and save the workflow for next time.