@tendrilapp/cli 0.1.39 → 0.1.41

package/dist/SKILL.md

@@ -7,8 +7,8 @@ description: MUST be used to implement, build, code up, recreate, or port a UI c
 
 Tendril's pipeline: RECORD the design's ground truth from Figma →
 GENERATE an implementation (you propose; Tendril's local ruler judges)
-→ VERIFY with per-config scores and evidence images → publish to the
-user's Tendril portal.
+→ VERIFY with per-config scores and evidence images → HAND OVER the
+bundle with its persisted report and evidence beside it.
 
 Non-negotiables (the CLI enforces these; do not fight them):
 - Only `tendril_engine_score` / `tendril verify` output counts as a
@@ -198,7 +198,10 @@ Batch runs (several components in one session):
    real faces is the only path to certification.
 3. Record each planned rep with FOUR tool calls: make the THREE
    Figma MCP calls in protocol order — get_metadata, then
-   get_design_context (excludeScreenshot=true), then get_screenshot —
+   get_design_context (excludeScreenshot=true), then get_screenshot
+   (ALWAYS with contentsOnly: true and maxDimension: 4096 — without
+   isolation the export bakes in the editor's component-set chrome,
+   which scores as ink no implementation can paint) —
    and then ONE `tendril_record_ingest_rep` carrying all three:
    metadata and context response text passed VERBATIM — responses
    often arrive as MULTIPLE output blocks (block counts vary BY
@@ -376,8 +379,24 @@ printed estimate.
 `tendril_verify` on any bundle directory recomputes everything —
 scores, behaviors, composition — and writes evidence images
 (render/ref/diff per config) next to the bundle. It is free and needs
-no account, always. Show the user the summary line and where the
-evidence lives.
+no account, always.
+
+It also PERSISTS its report to `verify-evidence/verify-report.json`,
+with machine-specific paths stripped so the directory can be handed to
+someone else as-is. Per config the report NAMES the evidence actually
+on disk — or `null` where a scoring path produced a score with no
+pixels, which is a real outcome and not a bug.
+
+The verdict, the pixels and the diff-colour legend join into one offline
+sheet at `verify-evidence/inspect.html`. That sheet comes from the CLI
+and has no MCP tool of its own, so OFFER to run
+`npx @tendrilapp/cli inspect <bundle-dir>` rather than asking the user
+to type it, and say what it produces before you run it.
+
+Point the user at that sheet, not at terminal scrollback: a verdict
+nobody can re-read beside the evidence it describes cannot be handed to
+anyone. Note the ORDER — `inspect` writes crops into the evidence
+directory it documents, so run it after `verify`, never before.
 
 NEVER re-score a bundle against a DIFFERENT recording set than the one
 it is bound to — scoring rewrites the bundle's verification identity

package/dist/tendril-mcp.js

@@ -88,10 +88,24 @@ var TOOLS = [
   {
     name: "tendril_doctor",
     annotations: { readOnlyHint: true },
-    description: "Machine readiness + version status in one shot: installed vs latest version (with publish date and update remediation), browser identity, font-cache state, Figma desktop MCP reachability. Use to self-diagnose before recording/scoring, or whenever versions are in question. Exit 1 = something not ready; the report says exactly what and how to fix it.",
+    description: "Machine readiness + version status in one shot: installed vs latest version (with publish date and update remediation), browser identity, font-cache state, Figma desktop MCP reachability, and whether this machine holds a LIVE portal session (checked against the portal itself; informational \u2014 publishing needs it, verification never does; when absent, offer tendril_login). Use to self-diagnose before recording/scoring, or whenever versions are in question. Exit 1 = something not ready; the report says exactly what and how to fix it.",
     schema: z.object({}),
     argv: () => ["doctor"]
   },
+  {
+    name: "tendril_login",
+    description: "Connect this machine to the user's Tendril portal account WITHOUT the user touching a terminal \u2014 phase one of the browser-approve sign-in. Starts the handshake, opens their browser to the approve page, and returns a verification link plus a short code. RELAY BOTH to the user verbatim: they click Approve on the page, and they must check the page shows EXACTLY this code (that check is the phishing defence \u2014 an attacker can send someone else's approve link). Then call tendril_login_wait to finish. Offer this whenever tendril_doctor reports no portal session and the user wants to publish or share; publishing needs it, verification never does.",
+    schema: z.object({
+      portal: optStr("portal origin override for self-hosted portals (defaults to https://app.trytendril.com)")
+    }),
+    argv: (i) => ["login", "--device-start", ...typeof i["portal"] === "string" ? ["--to", i["portal"]] : []]
+  },
+  {
+    name: "tendril_login_wait",
+    description: "Phase two of the browser-approve sign-in: waits (up to ten minutes, with live progress) for the user to click Approve on the page tendril_login opened, then stores the session on this machine. Call it right after relaying the link and code. A denial, a lapse, and success each come back as their own sentence; confirm the result to the user, and on success they are ready to publish.",
+    schema: z.object({}),
+    argv: () => ["login", "--device-wait"]
+  },
   {
     name: "tendril_record_next",
     annotations: { readOnlyHint: true },
@@ -112,7 +126,7 @@ var TOOLS = [
   },
   {
     name: "tendril_record_ingest_rep",
-    description: "Ingest a rep's ENTIRE recording in ONE call \u2014 the get_metadata response, the get_design_context response, and the get_screenshot image_url together. Make the three Figma calls first, in protocol order (get_metadata, then get_design_context with excludeScreenshot=true, then get_screenshot), then pass all three here VERBATIM. PREFER THIS over three separate ingest/fetch calls: one approvable operation per rep instead of three. Pieces land independently: on a partial failure the error names exactly which piece(s) to re-record \u2014 the rest are already on disk. The response carries `next` and, for design context, `assets` (auto-fetched server-side; only listed failures need record_asset).",
+    description: "Ingest a rep's ENTIRE recording in ONE call \u2014 the get_metadata response, the get_design_context response, and the get_screenshot image_url together. Make the three Figma calls first, in protocol order (get_metadata, then get_design_context with excludeScreenshot=true, then get_screenshot with contentsOnly=true and maxDimension=4096 \u2014 isolation keeps the editor's component-set chrome out of the padded export, where it would otherwise score as unpaintable reference ink), then pass all three here VERBATIM. PREFER THIS over three separate ingest/fetch calls: one approvable operation per rep instead of three. Pieces land independently: on a partial failure the error names exactly which piece(s) to re-record \u2014 the rest are already on disk. The response carries `next` and, for design context, `assets` (auto-fetched server-side; only listed failures need record_asset).",
     schema: z.object({
       setDir: str("recording set directory"),
       rep: str("planned rep slug"),