microsoft/diagnose-deployment
>- Surfaces PAC CLI upload errors and Dataverse async operation errors, pattern-matches against a known failure catalog, and optionally auto-fixes identified issues. Use when deployment errors", "fix deployment issues", "show upload logs", "why did my deploy fail", or "troubleshoot upload".
npx skills add https://github.com/microsoft/power-platform-skills --skill diagnose-deployment
> Plugin check: Run node "${PLUGIN_ROOT}/scripts/check-version.js" — if it outputs a message, show it to the user before proceeding.
Surfaces and pattern-matches deployment errors against a known failure catalog. For each identified error with an available auto-fix, asks explicit user permission before applying any changes. Never auto-applies fixes without confirmation.
powerpages.config.jsonCreate all tasks upfront at the start of this phase.
Tasks to create:
Steps:
powerpages.config.json in cwd and parent directoriespac --version (report version or "not installed")pac env who (report environment URL or "not authenticated")az account show (report subscription or "not logged in")Auth failures are non-blocking — report them as findings, continue collecting other artifacts.
> Reference: ${PLUGIN_ROOT}/references/alm-docs-grounding.md
Cap this step at ~30 seconds. If MCP search / fetch errors out, log a one-line note and continue — this skill must remain runnable offline.
microsoft_docs_search with the query: Power Pages deployment errors solution import troubleshooting.https://learn.microsoft.com/en-us/power-platform/alm/solution-concepts-alm (and at most one sister page on troubleshooting or known import errors) in parallel via microsoft_docs_fetch.${PLUGIN_ROOT}/references/deployment-error-catalog.md and flag any new error patterns not yet captured in the catalog.Gather all available context:
powerpages.config.json — extract siteName, websiteRecordId, compiledPath.powerpages-site/ folder exists.powerpages-site/*-manifest.yml — list all found, note their environment hostnames.solution-manifest.json exists (for solution-related diagnostics)docs/alm/last-import.json exists (for recent import failures){compiledPath}/ exists and is non-emptyReport: "Found project: {siteName}. Artifacts collected."
Re-run pac pages upload-code-site in capture mode to get fresh error output:
pac pages upload-code-site --rootPath "." 2>&1
> Note: This intentionally triggers the upload to capture any errors. If the upload succeeds cleanly, that is also a valid diagnostic result ("no errors found").
Capture stdout+stderr as a single string. Pass to scripts/parse-deployment-errors.js:
echo "{escaped-output}" | node "${PLUGIN_ROOT}/scripts/parse-deployment-errors.js"
Parse the JSON findings array. If the upload succeeded with no errors, note this and skip to Phase 5 with an empty findings list.
Only run if .solution-manifest.json exists.
GET {envUrl}/api/data/v9.2/asyncoperations?$filter=statecode eq 3 and statuscode eq 31 and createdon gt {yesterday}&$select=asyncoperationid,name,message,friendlymessage,statuscode,completedon&$orderby=completedon desc&$top=5
GET {envUrl}/api/data/v9.2/importjobs?$select=solutionname,completedon,progress&$orderby=completedon desc&$top=3
message field through parse-deployment-errors.jsSkip gracefully if auth is not available (auth failure in Phase 1).
Consolidate all findings from Phases 3 and 4. For each finding, categorize:
Also add findings for missing artifacts discovered in Phase 2:
websiteRecordId → Error (patternId: missing-website-record-id)empty-build)Present all findings in a table:
| # | Severity | Type | Issue | Auto-fix? |
|---|---|---|---|---|
| 1 | Error | upload | JavaScript uploads blocked | Yes |
| 2 | Warning | config | Multiple manifest files found | No |
For each Error finding with autoFixAvailable: true, in order:
<!-- gate: diagnose-deployment:6.auto-fix | category=consent | cancel-leaves=nothing -->
> 🚦 Gate (consent · diagnose-deployment:6.auto-fix): Per-finding consent before applying any auto-fix. Loops once per Error finding with autoFixAvailable: true — each finding gets its own Yes / No / Skip-all AskUserQuestion. The pattern ID surfaces in the prompt. Never batch fixes — three findings = three separate consent prompts (unless the user picks "Skip all" on the first, which short-circuits the loop). The Yes from finding 1 does NOT cover finding 2; each fix has its own blast radius (different files, different settings, different reversibility).
AskUserQuestion:> "Issue: {message}
> Proposed fix: {suggestedFix}
> Apply this fix? Yes / No / Skip all auto-fixes"
stale-manifest: Delete *-manifest.yml file(s) in .powerpages-site/
# Ask which manifest to delete if multiple found, then:
rm ".powerpages-site/{manifestFile}"
blocked-js: Update blockedattachments setting
# Get current setting first
pac env list-settings | grep -i blocked
# Remove .js from the blocked list (preserve other blocked types)
pac env update-settings --name blockedattachments --value "{updated-value}"
missing-website-record-id: Retrieve and update record ID
pac pages list
# Parse output, find matching site by name, then update powerpages.config.json
auth-expired: Guide re-authentication
pac auth create --environment "{envUrl}"
az login
empty-build: Run build
npm run build
> Key Constraint: Never apply any fix without explicit user permission. Each fix requires a separate confirmation.
Display a final summary table of all findings:
| # | Severity | Type | Issue | Status |
|---|---|---|---|---|
| 1 | Error | upload | JS uploads blocked | Fixed |
| 2 | Error | config | Missing websiteRecordId | Manual required |
| 3 | Warning | config | Multiple manifest files | Informational |
Status values:
If all errors are resolved: suggest retrying deployment with /power-pages:deploy-site.
If manual steps remain: list them explicitly with commands or links.
> Reference: ${PLUGIN_ROOT}/references/skill-tracking-reference.md
Follow the skill tracking instructions in the reference to record this skill's usage. Use --skillName "DiagnoseDeployment".
pac pages upload-code-site produces no output: report "No upload errors detected in current state"parse-deployment-errors.js returns no findings: report "No known error patterns detected" and show raw output for manual review| Task subject | activeForm | Description |
|---|---|---|
| Verify prerequisites and locate project | Verifying prerequisites | Check PAC CLI, Azure CLI, locate project root and powerpages.config.json |
| Collect deployment artifacts | Collecting deployment artifacts | Read config, list manifests, check build output, check solution manifest |
| Surface upload errors | Surfacing upload errors | Re-run pac pages upload-code-site in capture mode, parse stdout/stderr |
| Query solution import status | Querying solution import status | Check recent failed async operations and import jobs in Dataverse |
| Diagnose and categorize findings | Diagnosing findings | Pattern-match all errors against deployment-error-catalog, assign severity |
| Offer auto-fixes | Applying auto-fixes | For each fixable error, ask permission and execute fix, verify result |
| Present findings summary | Presenting summary | Show all findings table with severity and fix status, list manual steps |
Take microsoft/diagnose-deployment from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.