Add WebMCP interaction skill + SVO Zählerstandsmeldung domain skill

[VibeVik89]

What

Two new skills capturing field-tested learnings about WebMCP (navigator.modelContext) — page-registered, agent-callable tools.

interaction-skills/webmcp.md (generic)

• Detecting WebMCP (the API surface lives on the prototype: getTools, executeTool, registerTool, ontoolchange).
• Listing tools — inputSchema frequently arrives as a JSON-encoded string, not an object.
• The two non-obvious executeTool gotchas, verified against Chrome (June 2026):
  1. First arg must be the RegisteredTool object from getTools(), not the tool name → else TypeError: ... not of type 'RegisteredTool'.
  2. Second arg must be a JSON string, not a JS object → else UnknownError: Failed to parse input arguments.
• Verify-via-DOM, because handlers may intentionally not echo entered values.

domain-skills/svo/zaehlerstandsmeldung.md (site-specific)

• The report_meter_reading tool: required fields, the HT/NT-vs-single-rate conditional fields, the honeypot trap, and that it never submits (CAPTCHA stays a human action).

Why

The executeTool arg shape is undiscoverable from the API alone and cost several attempts to work out — the next agent on any WebMCP site shouldn't pay that tax again.

🤖 Generated with Claude Code

https://claude.ai/code/session_01EqyH3G9SuZmQXnotdt4bwi

---

Summary by cubic

Adds a generic WebMCP interaction skill and a site-specific SVO meter-reading skill to use page-registered tools instead of the DOM, reducing errors and setup time.

• New Features
  • interaction-skills/webmcp.md: Detects navigator.modelContext, lists tools (parses inputSchema when it’s a JSON string), and details executeTool requirements (pass the RegisteredTool object and a JSON string). Recommends DOM-based verification.
  • domain-skills/svo/zaehlerstandsmeldung.md: Documents report_meter_reading inputs, HT/NT vs single-rate rules, and the honeypot trap. Notes that the tool fills fields but never submits (CAPTCHA remains manual).

^{Written for commit 15f9fa6. Summary will update on new commits.}

[browser-harness-review]

✅ Skill review passed

Reviewed 2 file(s) — no findings.

[cubic-dev-ai]

3 issues found across 2 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="interaction-skills/webmcp.md">

<violation number="1" location="interaction-skills/webmcp.md:78">
P2: Verification snippet may leak sensitive form data by logging values from nearly all input fields (including non-hidden secrets).</violation>
</file>

<file name="domain-skills/svo/zaehlerstandsmeldung.md">

<violation number="1" location="domain-skills/svo/zaehlerstandsmeldung.md:62">
P2: Syntax error: missing closing parenthesis for `print()` in the verify code example. The expression `print(js(r'''...''')` has an unclosed `print(` call.</violation>

<violation number="2" location="domain-skills/svo/zaehlerstandsmeldung.md:69">
P3: The note advises reading `inputSchema.required`, but per the companion skill (`interaction-skills/webmcp.md`) the `inputSchema` is **often a JSON-encoded string, not an object**. Accessing `.required` directly on a string returns `undefined`. The note should include a `JSON.parse()` step before reading `.required` so the guidance works in the common case.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Fix all with cubic | Re-trigger cubic</sub>

[cubic-dev-ai]

P2: Verification snippet may leak sensitive form data by logging values from nearly all input fields (including non-hidden secrets).

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At interaction-skills/webmcp.md, line 78:

<comment>Verification snippet may leak sensitive form data by logging values from nearly all input fields (including non-hidden secrets).</comment>

<file context>
@@ -0,0 +1,92 @@
+res = js(r'''(() => {
+  const out = [];
+  document.querySelectorAll('input,select,textarea').forEach(el => {
+    if (el.type === 'hidden') return;
+    out.push({name: el.name || el.id, value: (el.type==='checkbox'||el.type==='radio') ? el.checked : el.value});
+  });
</file context>

[cubic-dev-ai]

P2: Syntax error: missing closing parenthesis for print() in the verify code example. The expression print(js(r'''...''') has an unclosed print( call.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At domain-skills/svo/zaehlerstandsmeldung.md, line 62:

<comment>Syntax error: missing closing parenthesis for `print()` in the verify code example. The expression `print(js(r'''...''')` has an unclosed `print(` call.</comment>

<file context>
@@ -0,0 +1,69 @@
+    out.push({name: el.name, value: (el.type==='checkbox'||el.type==='radio') ? el.checked : el.value});
+  });
+  return out;
+})()''')
+```
+
</file context>

[cubic-dev-ai]

P3: The note advises reading inputSchema.required, but per the companion skill (interaction-skills/webmcp.md) the inputSchema is often a JSON-encoded string, not an object. Accessing .required directly on a string returns undefined. The note should include a JSON.parse() step before reading .required so the guidance works in the common case.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At domain-skills/svo/zaehlerstandsmeldung.md, line 69:

<comment>The note advises reading `inputSchema.required`, but per the companion skill (`interaction-skills/webmcp.md`) the `inputSchema` is **often a JSON-encoded string, not an object**. Accessing `.required` directly on a string returns `undefined`. The note should include a `JSON.parse()` step before reading `.required` so the guidance works in the common case.</comment>

<file context>
@@ -0,0 +1,69 @@
+
+- There's a `honeypot` text input — leave it empty (spam trap; filling it likely flags the submission).
+- The page has a fixed notification banner overlaying the lower viewport; `scrollIntoView`/`window.scrollTo` to bring form fields above it before screenshotting.
+- Conversational use: read the tool's `inputSchema.required` first, then ask the user for the tariff type + the matching reading(s) before calling.
</file context>