Map, analyze, and improve your HubSpot CRM.

Continue in the same agent workspace from the completed Setup verification. Reuse the account, portal ID, and CLI readiness already confirmed there. Do not repeat installation, authentication, skills installation, or `hubspot whoami`. If Setup verification did not finish with an overall `Ready` status, stop and tell me to ask a workshop facilitator for help.

Use the HubSpot Agent CLI and any installed HubSpot Agent CLI Skills to create a focused, read-only account map. If you are unsure about a command, inspect `hubspot --help` instead of guessing.

Build the map:

1. Discover the CRM object types available to this account.
2. Choose up to five business-relevant objects for the map. Prioritize accessible objects that contain data and include up to two custom objects when available. Adapt to this account instead of assuming which objects or HubSpot products it uses.
3. For each selected object, report its display name, internal type, record count or "not verified," and two or three notable properties based only on returned metadata. Identify standard or custom objects only when the CLI provides that evidence.
4. Verify up to three meaningful relationships between the selected objects. Prefer relationship metadata supported by the current authentication. If a metadata command rejects the current OAuth credentials, do not bypass it or ask for another credential. Use the smallest read-only record sample needed to check real associations, retrieving only record IDs and association metadata—not property values. State the evidence method and sample size. Treat an inconclusive check as "not verified."
5. Create a concise visual map in a self-contained file named `hubspot-account-map.html`. Show every selected object as a node, even when no relationship is verified. Draw no more than three relationship edges, and draw an edge only when the CLI returned supporting evidence. If you cannot create local files, return the same map as a Markdown table and text-based relationship diagram.

Follow these guardrails:

- Use only read operations through the Agent CLI. Do not change or preview changes to HubSpot.
- Prefer metadata and record counts over record retrieval. Do not export complete record sets or display sensitive property values, credentials, or the authenticated user's email in the artifact.
- Treat inaccessible information as "not verified," not absent. Continue with whatever evidence the account and permissions make available.
- Do not infer subscriptions or product adoption from object availability alone.
- Stop after documenting the current state; do not recommend changes yet.

Your HubSpot Account Map must include:

1. The account name or label already confirmed during setup, if available, and the portal ID—without rerunning a connection check.
2. A focused object inventory with record-count status and property highlights.
3. A visual map containing every selected object and only verified relationship edges.
4. A "What I could not verify" section for permissions, evidence gaps, and the relationship-verification method used.
5. Confirmation that you used read-only Agent CLI operations and made no CRM changes.

When the map is complete, show me the file location if you created one, give me a three-bullet summary, and wait for the next workshop step.

Continue from the HubSpot Account Map you created in Step 1. Analyze that verified current state and recommend practical improvements for this account.

Remain read-only. Do not create, edit, delete, merge, import, associate, enroll, schedule, or preview any changes in HubSpot. You may use up to three additional targeted read-only Agent CLI data checks only when they resolve an important evidence gap. A help lookup for command syntax does not count as a data check. Before each data check, name the uncertainty it will resolve, and stop checking once the available evidence is sufficient. Prefer metadata, counts, and aggregate results; do not expose sensitive record values.

Create up to three supported recommendations. Fewer than three is correct when the evidence does not support more. If no change is well supported, recommend a read-only validation step instead of inventing a problem.

1. Start with a specific observation from the account map.
2. Separate verified evidence from interpretation. Label incomplete evidence as an assumption or "not verified."
3. Explain the likely business or operational impact in plain language. If the impact depends on unverified business intent, label it as potential—not confirmed.
4. Recommend one proportionate improvement. Prefer a read-only validation step when the underlying gap or business need is not yet verified. A read-only validation step may use ad hoc Agent CLI queries and return results in chat or a local file; it must not create or save a HubSpot report, list, view, dashboard, workflow, or other portal asset.
5. Rate its likely impact and implementation effort as Low, Medium, or High.
6. State the confidence level and what someone should verify before acting. Do not assign High confidence when the conclusion depends on unverified business intent.

Use these evidence rules:

- An empty object, zero count, or missing association is not automatically a gap; it may be intentional.
- Do not infer subscriptions, feature adoption, team behavior, business model, intended process, or business need from an object or property name, availability, or count alone.
- Call something a gap only when verified account evidence indicates that an intended outcome or existing process is not being met.
- Put a potentially useful but unverified idea under "Open questions," not in the ranked recommendations. Include no more than two open questions.
- It is valid to conclude that the checked areas appear healthy.

Adapt the analysis to the account. Consider data quality, CRM structure, lifecycle or process consistency, relationship coverage, and reporting readiness only when the map provides relevant evidence. If this is a sparse demo account, recommend foundational setup or validation steps instead of inventing problems.

Rank only the supported recommendations, using this order:

- Strongest evidence of a real problem or useful measurement first.
- Higher likely impact first when evidence strength is similar.
- Lower effort first when evidence strength and impact are similar.

Update `hubspot-account-map.html` with a new section named "Analysis and recommendations." Preserve the existing map. Add a compact priority table containing: rank, verified observation, evidence, likely impact, recommendation, effort, confidence, and verification needed. If conditional opportunities remain, place up to two of them in a separate "Open questions" list outside the ranking. If no local file exists, return the section in Markdown instead.

When complete, show me the updated file location, summarize the highest-priority supported recommendation or validation step in two sentences, confirm that you made no CRM changes, and wait for the next workshop step.

Continue from the Account Map and recommendations you completed in Steps 1 and 2. Choose the highest-priority recommendation that is both useful and sufficiently supported by evidence, then design one safe follow-up workflow someone could validate after this workshop. Reuse only the evidence already present in the artifact; do not run any new Agent CLI or other data checks during this step.

This is a planning exercise only. Do not create or schedule an automation. Do not create, edit, delete, merge, import, associate, enroll, or preview changes in HubSpot. Do not assume that a feature, permission, integration, or subscription is available; mark each prerequisite as "verified" or "needs verification." Preserve the evidence labels from Step 2: do not turn a conditional opportunity into a confirmed problem or infer a business model, intended process, or operational purpose from object names, counts, or relationships alone.

Design the workflow with these parts:

1. Outcome: the practical result the workflow should produce.
2. Starting point: the schedule, event, file, or human request that would begin it.
3. Read-only inputs: the minimum HubSpot metadata or records it would need.
4. Decision logic: how it would turn those inputs into a useful result.
5. Output: a report, draft, alert, or proposed action. Default to an output outside HubSpot.
6. Human checkpoint: what a person must review before any CRM write or external communication.
7. Guardrails: scope limits, privacy considerations, failure conditions, and how to prevent unsupported assumptions.
8. Prerequisites: required access, tools, data, or HubSpot capabilities, each labeled "verified" or "needs verification."
9. Success signal: one observable way to decide whether the workflow is useful. It must remain valid when the result shows no problem; success cannot require finding a defect or opportunity.

Classify every operation the workflow actually proposes as one of the following. Do not add setup checks, discovery commands, or hypothetical operations just to populate the classification:

- READ: retrieves information without changing HubSpot.
- DRAFT: prepares an output or proposed action for human review.
- WRITE: would change HubSpot or contact someone and therefore requires separate approval outside this workshop.

Create a reusable starter prompt for a future test run. The starter prompt must default to READ and DRAFT operations, begin with a dry run, stop before any WRITE operation, and require explicit human confirmation before proceeding beyond the plan.

Update `hubspot-account-map.html` with a final section named "What comes next." Include the workflow outline, operation classifications, prerequisites, guardrails, success signal, and reusable starter prompt. Preserve the existing sections and add exactly one "What comes next" section. Before finishing, perform a local privacy sweep of the complete artifact: remove the authenticated user's email and ensure no credentials, sensitive property values, or unnecessary personal data appear. If no local file exists, return the section in Markdown instead.

Finish with a concise, factual recap of what the account map revealed, the recommended priority, and the proposed follow-up workflow. Preserve all "potential," "not verified," and conditional labels; do not describe an inferred business process as fact. Confirm that you ran no new data checks, did not create or schedule an automation, and made no CRM changes.