/migrate discover — Discovery Interview

Conduct or resume the infrastructure discovery interview for the current migration assessment. Walks through the discovered dimensions of infrastructure discovery with branching questions. Supports confirmation-first mode for dimensions pre-filled by /migrate new.

Instructions

1. Load State

1. Call the MCP tool get_assessment with project_path set to the current working directory. If it returns null, fall back to reading .migration/assessment.json. If neither exists, tell the user to run /migrate new first.
2. Call the MCP tool get_discovery with the assessment ID (no dimension filter) to get all existing discovery data. If it returns null, fall back to scanning .migration/discovery/ for existing dimension files.
3. Load the discovery tree using a cascading fallback:
   • Primary: Call the MCP tool get_composed_discovery_tree with the assessment_id. This automatically resolves all knowledge packs from the assessment's source_stack/target_stack, merges dimensions by ID (higher-priority packs win), and returns a single deduplicated dimensions array with packs_used metadata.
   • Fallback 1: If the composed tree returns empty or fails, call get_discovery_tree with pack_ids derived from the assessment's source_stack.platform (e.g., ["sitecore-xp"]). If source_stack is empty, use ["sitecore-xp"] as default.
   • Fallback 2: If MCP tools return null or empty dimensions, fall back to reading skills/migrate-knowledge/discovery/discovery-tree.json.
4. Read dimension descriptions from skills/migrate-knowledge/discovery/dimension-descriptions.md for context.

2. Show Progress

Display the current discovery progress using four dimension states (use the actual count of dimensions from the loaded discovery tree):

Discovery Progress: 3 of N dimensions complete, 5 pre-filled

  [x] Compute/Hosting       5/7 confirmed
  [~] Database               6/8 assumed (from overview — needs review)
  [~] Search                 4/7 assumed (from overview — needs review)
  [~] Caching                3/6 assumed (from overview — needs review)
  [~] CDN                    2/6 assumed (from overview — needs review)
  [ ] DNS
  [/] SSL/TLS                1/5 partial (interview started)
  [ ] Storage/Media
  ...

Dimension state markers:

• [x] Complete — All required questions have confidence: "confirmed" answers
• [~] Pre-filled — Has assumed answers from overview extraction, needs user review. Detected when: the JSON snapshot has "source": "overview_extraction", OR all answers have confidence: "assumed" with basis values starting with "extracted from", "inferred:", or "domain knowledge:"
• [/] Partial — Interview was started but not finished (mix of confirmed and unanswered questions, or user interrupted mid-dimension)
• [ ] Not started — No answers recorded

For pre-filled [~] dimensions, show the count annotation with "(from overview — needs review)" to make it clear these need confirmation. For [x] and [/] dimensions, show the count of answered questions.

Recommended order: Suggest starting with [~] pre-filled dimensions first (confirmation is faster than fresh interviews), then [/] partial dimensions, then [ ] not-started dimensions.

3. Conduct Interview

For each incomplete dimension, in order:

Detecting Pre-filled Dimensions

Before starting a dimension, check if it has pre-filled answers. A dimension is pre-filled if:

• The JSON snapshot contains "source": "overview_extraction", OR
• All existing answers have confidence: "assumed" with basis values matching extraction patterns ("extracted from initial overview", "inferred: ...", "domain knowledge: ...")

If pre-filled, use Confirmation-First Mode (below). Otherwise, use the standard interview flow.

Confirmation-First Mode (for pre-filled dimensions)

When entering a dimension with pre-filled answers:

Step 1 — Present summary of what's captured. Show a compact list of all assumed answers for this dimension:

Database — 6 of 8 required questions pre-filled from your overview:

  Engine:           Amazon RDS for SQL Server  (assumed)
  SQL Version:      SQL Server 2019            (assumed)
  Instance Class:   db.m5.xlarge               (assumed — not in overview, inferred from instance sizing)
  HA Config:        RDS Multi-AZ               (assumed)
  Encryption:       Yes                        (assumed — domain knowledge: standard for production RDS)
  Separate xDB:     No                         (assumed)

  Still needed:
  - Total database size (GB)
  - Custom databases beyond standard Sitecore DBs

Step 2 — Accept bulk confirmation. Ask: "Does this look right, or do you want to correct anything?"

• If the user says "looks good", "yes", "correct", or similar → promote ALL assumed answers in this dimension to confidence: "confirmed", update their basis to "confirmed by user (originally: <original basis>)", then proceed to ask only the gap questions (Step 4)
• If the user says "looks good except..." or provides partial corrections → handle corrections first (Step 3), then bulk-confirm the rest

Step 3 — Handle corrections. If the user corrects a value:

• Update the corrected answer with confidence: "confirmed" and basis: "corrected by user"
• Check if the correction invalidates any inferred values in this or other dimensions. For example, if the user corrects the database engine from RDS to "SQL Server on EC2", re-evaluate inferences that depended on RDS (like database_ha_config: "RDS Multi-AZ")
• If dependent inferences are invalidated, remove them and add those questions to the gap list
• After corrections, bulk-confirm remaining uncorrected assumed answers

Step 4 — Ask only about gaps. Transition to asking only the unanswered required questions and any newly-ungated conditional questions. Use the same conversational style as the standard interview flow.

Step 5 — Save with mixed confidence. When saving:

• Confirmed answers (bulk or individual) get confidence: "confirmed"
• Corrected answers get confidence: "confirmed" with basis: "corrected by user"
• Any assumed answers the user explicitly left alone get confidence: "confirmed" (they reviewed and accepted them)
• Answers to gap questions get confidence: "confirmed"

Standard Interview Flow (for non-pre-filled dimensions)

Starting a Dimension

• Announce which dimension you're starting and briefly explain what it covers (use dimension-descriptions.md)
• Ask the required questions for that dimension conversationally
• You don't have to ask questions one-by-one — group related questions naturally
• Accept partial answers; mark unknown items with "confidence": "unknown"

Branching Logic

• After getting answers to required questions, check conditional questions
• Only ask conditional questions whose conditions are met based on previous answers
• Example: If search.type == 'solr_cloud', ask about ZooKeeper configuration

Inference Rules

• Apply inference rules from the discovery tree when answers aren't provided
• Tell the user what you're inferring: "Since you're on Sitecore 10.3, I'll assume Windows Server 2019 or 2022 — correct?"
• Record inferences with their confidence level

Handling "I Don't Know"

• Accept "I don't know" gracefully — record as "confidence": "unknown"
• Note what impact this gap has: "No problem — I'll flag this as a gap. It affects estimate accuracy for the database phase by approximately ±20%."
• Let the user know that assumed and unknown answers will be formally tracked: "I'll record this as an assumption. When you run /migrate analyze, it'll be registered in the assumptions tracker with its impact on the estimate range and a suggested way to validate it."

Saving Progress

After completing each dimension (or when the user wants to stop):

Step 1 — Save to MCP (primary): Call the save_discovery MCP tool with the assessment ID, dimension name, status, and all answers (including inferred values with their basis). Inferred values should be included in the answers object with confidence set to "assumed" and the basis field explaining the inference rule.

Step 2 — Write JSON snapshot: Write .migration/discovery/<dimension>.json with the same data:

{
  "dimension": "compute",
  "status": "complete",
  "completed_at": "2026-02-20T...",
  "answers": {
    "compute_cm_instances": {
      "value": 1,
      "notes": "",
      "confidence": "confirmed"
    },
    "compute_cd_instances": {
      "value": 4,
      "notes": "Auto-scales to 6 during peaks",
      "confidence": "confirmed"
    }
  },
  "inferred": {
    "infer_os_version": {
      "value": "Windows Server 2019",
      "basis": "Sitecore 10.3 requires Windows Server 2019+",
      "confidence": "medium"
    }
  }
}

Set "status": "partial" if not all required questions were answered.

When saving a previously pre-filled dimension:

• Remove the "source": "overview_extraction" field from the JSON snapshot (the dimension has now been reviewed)
• Set status to "complete" only when ALL required questions have confidence: "confirmed" answers
• If the user reviewed and confirmed everything, all answers should be "confirmed" and status should be "complete"
• If the user skipped some gap questions, status remains "partial"

4. Interruption Handling

The user can interrupt at any time. When they do:

• Save progress for the current dimension (even if partial)
• Update .migration/assessment.json with "updated": "<now>"
• Show progress summary
• Tell them they can resume with /migrate discover

5. Completion

When all dimensions are complete (or the user indicates they're done):

• Show a summary of discovery completeness
• Highlight any dimensions marked as partial or not_started

Call out dimensions that still have only assumed answers (pre-filled but never reviewed by the user):

Note: 3 dimensions have only assumed answers (never reviewed):
  [~] DNS           — 4 assumed answers from overview extraction
  [~] SSL/TLS       — 2 assumed answers from overview extraction
  [~] Monitoring    — 1 assumed answer from overview extraction

These will be treated as assumptions in the estimate, widening the
confidence range. Run /migrate discover again to review them.

• Update .migration/assessment.json status to "analysis" if all dimensions are at least partial
• Suggest next step: /migrate analyze for cross-reference analysis, or /migrate gaps to review unknowns

Interview Style

• Be conversational but efficient — don't read questions robotically
• Group related questions: "Let's talk about your database setup. What SQL Server version is running, and is it RDS or self-managed on EC2? How many databases do you have roughly?"
• Acknowledge answers and provide context: "4 CD instances with auto-scaling to 6 — that's a moderately scaled deployment. This will trigger some complexity multipliers in the estimate."
• When you spot risk patterns emerging, mention them casually: "Interesting — SolrCloud with SXA. I'll flag a potential caching interaction risk when we get to analysis."
• If the user provides information out of order, accept it and file it in the right dimension
• If the user dumps a lot of info at once (like pasting infrastructure docs), parse it all and fill in multiple dimensions
• In confirmation-first mode, be concise — the user has already provided this info once, don't make them re-read lengthy explanations

Dimension Order

Process dimensions in this order (infrastructure-up):

1. Compute/Hosting
2. Database
3. Search (Solr)
4. Caching (Redis)
5. CDN
6. DNS
7. SSL/TLS
8. Storage/Media
9. Email (EXM)
10. xConnect/xDB
11. Identity
12. Session State
13. Custom Integrations
14. CI/CD
15. Monitoring
16. Networking/Firewall
17. Backup/DR

The user can request to skip to a specific dimension or do them out of order.

When pre-filled dimensions exist, suggest reviewing [~] dimensions first (faster confirmation flow), then [/] partial dimensions, then [ ] not-started dimensions.

Post-Completion Suggestion

After all dimensions are complete and data is saved, suggest running a challenge review:

Discovery complete! Before moving to analysis, consider running /migrate challenge discovery to validate data quality, check for inconsistencies across dimensions, and verify Azure target services are still current. This is optional but recommended for high-confidence estimates.