Plugin check: Run node "${CLAUDE_PLUGIN_ROOT}/scripts/check-version.js" — if it outputs a message, show it to the user before proceeding.

Scan Site

Run a security scan on a deployed Power Pages site, fetch the latest scan report, and surface findings in a plain-language summary. The scan runs server-side; duration depends on site size — small sites finish in minutes, large sites can take hours.

This skill scans the live deployed site, not local source code.

Initial request: $ARGUMENTS

Gotchas

• Website record id vs portal id. .powerpages-site/website.yml stores the website record id, not the portal id. Every script takes --portalId. Resolve once via website.js --websiteId during prerequisites.
• Never resolve by name. Site names can duplicate inside an environment; only the website record id is safe.
• null from the resolver means the site is not deployed, or the authenticated profile points at a different environment.
• Scans are long-running. Duration depends on site size — small sites finish in minutes, large sites can take hours. Poll in the background and increase --timeoutMinutes for large sites.
• Only one scan per site at a time. A start while a scan is running returns Z003 — start-deep-scan.js reports it as { "status": "already-running" } (exit 0).
• Rate limits may apply. The service may throttle repeated scans on the same site. When throttled, wait and retry later.
• No completed scan yet. A fresh site or a site mid-scan has no completed report — get-latest-report.js returns { "status": "empty" }.

Workflow

1. Prerequisites — Locate project, confirm sign-in, identify site
2. Check scan state — Detect whether a scan is currently running
3. Choose an action — Context-aware recommendation (run new scan / show latest)
4. Run the scan — Start and poll for completion
5. Fetch and summarize — Get the report, present findings
6. Walk through follow-ups — Route issues to the right downstream skill (only if the report contains issues)

Task Tracking

Create tasks in four groups. Mark each in_progress when starting, completed when done.

1. Prerequisites

1.1 Locate the project, detect review mode

Use Glob to find **/powerpages.config.json. If $ARGUMENTS contains --review <out-dir>, remember the output directory — Step 3 (choose an action) is skipped, Step 4 (run scan) executes automatically (start a fresh scan or attach to a running one), Step 5 writes JSON only, and Step 6 (follow-ups) is skipped.

1.2 Resolve site identifiers

Read .powerpages-site/website.yml → extract id field → that is <WEBSITE_ID>.

If missing, the site has not been deployed. Tell the user and recommend /deploy-site. Stop. Do not resolve by name or URL.

Resolve to portalId:

node "${CLAUDE_PLUGIN_ROOT}/scripts/website.js" --websiteId "<WEBSITE_ID>"

Capture Id (portalId), Type, Name, WebsiteUrl. If exit code 2 → sign-in required (pac auth create or az login). If null → site not found in this environment. Stop in either case.

2. Check scan state

node "${CLAUDE_PLUGIN_ROOT}/skills/scan-site/scripts/poll-deep-scan.js" --portalId "<PORTAL_ID>" --once

--once does a single status check, exits 0, and prints:

• { "status": "ongoing" } → a scan is currently running.
• { "status": "idle" } → no scan running.

Then call get-latest-report.js to know whether a completed report exists:

node "${CLAUDE_PLUGIN_ROOT}/skills/scan-site/scripts/get-latest-report.js" --portalId "<PORTAL_ID>"

{ "status": "ok" } means a report is available. { "status": "empty" } means no completed scan exists.

3. Choose an action

Skip in review mode — go straight to Step 4 (which always runs in review mode).

MUST use plain language only. Never use words like CSP, CORS, OWASP, hardening, or scan profile.

Default approach

Analyze the site's current state and recommend the single most relevant action. Present the recommendation with a plain-language explanation of why. The user can accept or choose differently.

• Scan running, no completed report → recommend waiting for the running scan to finish.
• Scan running, report exists → recommend showing the latest results while the new scan continues.
• Idle, no completed report → recommend running a new scan.
• Idle, recent report exists → ask whether to use the existing report or run a fresh scan.

If the site's state does not warrant a specific recommendation, do not force one — ask what the user wants to do.

Option rules

When presenting options via AskUserQuestion:

• Keep label to 1–5 words. Include description on every option.
• For options that trigger a new scan, surface the relevant caveats inside that option's description so the user has them at decision time. Do not ask a separate confirmation question after the user picks the option.
• Include preview only when the option represents a concrete change (starting a new scan). Do not add preview to "show latest" or informational choices.
• Only show options that are actionable given the current state. If a scan is already running, do not offer "Start a new scan".
• Mark "(Recommended)" only when the site's state justifies it.

4. Run the scan

In review mode, always execute this step: if a scan is already running, attach to it and poll; otherwise start a fresh scan and poll. Do not ask — review mode runs end-to-end without user interaction.

In interactive mode, skip if the user chose to view the latest results.

Start the scan:

node "${CLAUDE_PLUGIN_ROOT}/skills/scan-site/scripts/start-deep-scan.js" --portalId "<PORTAL_ID>"

If stdout is { "status": "already-running" }, skip ahead to polling — there is already a scan in progress.

Then poll for completion:

node "${CLAUDE_PLUGIN_ROOT}/skills/scan-site/scripts/poll-deep-scan.js" --portalId "<PORTAL_ID>"

Run polling with run_in_background: true so the user can keep working. The script exits when the scan finishes or the timeout passes (default 20 minutes). If it times out, fetch whatever report is available and note the timeout in the summary.

5. Fetch and summarize

5.1 Fetch and transform the report

node "${CLAUDE_PLUGIN_ROOT}/skills/scan-site/scripts/transform-report.js" --portalId "<PORTAL_ID>"

Parse the stdout JSON. The status field can be:

• ok — a normal report with findings and details.
• empty — no completed scan exists for this site (e.g., fresh site or scan still running). Record a single info finding explaining this and continue.
• malformed — the API returned a response missing the Rules array. The transform emits a single warning finding describing this; surface it to the user and recommend re-running the scan.

See references/scan-reference.md for the Risk → severity mapping the script applies.

5.2 Review mode

In review mode, skip the HTML report and write the transform stdout to <REVIEW_DIR>/scan-site.json. Then stop. The transform emits { status, findings, details }; the orchestrating skill handles presentation.

5.3 Render HTML report

Skip in review mode.

Render uses the same shared template as the consolidated security review. Build a single-section review-data payload, then render:

node "${CLAUDE_PLUGIN_ROOT}/scripts/build-review-data.js" \
  --reportName "Site Scan" \
  --inpu