@ateam-ai/mcp 0.4.74 → 0.4.76

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@ateam-ai/mcp",
-  "version": "0.4.74",
+  "version": "0.4.76",
   "mcpName": "io.github.ariekogan/ateam-mcp",
   "description": "A-Team MCP Server — build, validate, and deploy multi-agent solutions from any AI environment",
   "type": "module",

package/src/tools.js

@@ -239,47 +239,75 @@ function _widgetHasRender(r) {
 //
 // Detected here rather than left to a person noticing an empty panel, and
 // reported with the repair attached — a signal the reasoning engine can act on.
+// Extract the FULL argument object of each postMessage(...) call by matching
+// braces, so every call is judged on its own text. The previous version took a
+// fixed 400-character window and concatenated all of them: a send object longer
+// than the window was inspected in part, and `source:"adas-plugin"` in one call
+// could make an unrelated postMessage elsewhere on the page be judged as the
+// ADAS protocol. Both were rejected in review, correctly — an approximate
+// boundary is not a structural one.
+function _postMessageObjects(html) {
+  const out = [];
+  const re = /postMessage\s*\(\s*\{/g;
+  let m;
+  while ((m = re.exec(html)) !== null) {
+    const open = html.indexOf("{", m.index);
+    let depth = 0, i = open, inStr = null, esc = false;
+    for (; i < html.length; i++) {
+      const c = html[i];
+      if (inStr) {
+        if (esc) { esc = false; continue; }
+        if (c === "\\") { esc = true; continue; }
+        if (c === inStr) inStr = null;
+        continue;
+      }
+      if (c === '"' || c === "'" || c === "`") { inStr = c; continue; }
+      if (c === "{") depth++;
+      else if (c === "}") { depth--; if (depth === 0) { i++; break; } }
+    }
+    // Unbalanced (minified truncation, template weirdness) → skip rather than
+    // guess. A message we cannot delimit is one we must not judge.
+    if (depth === 0) out.push(html.slice(open, i));
+    re.lastIndex = Math.max(re.lastIndex, i);
+  }
+  return out;
+}
+
 export function _widgetProtocolProblems(html) {
   if (typeof html !== "string" || !html.includes("postMessage")) return [];
 
-  // STRUCTURE, NOT TOKENS. The first version matched /correlationId/ anywhere in
-  // the file, so a correct widget with an unrelated local named correlationId was
-  // reported broken — and `fix_with` would then steer an agent into editing
-  // working code. Reviewer was right to reject it. Each check below anchors to
-  // the actual send or receive shape.
   const problems = [];
-
-  // Only look at what is POSTED to the host: postMessage(...) argument objects.
-  const sends = [...html.matchAll(/postMessage\s*\(\s*\{[\s\S]{0,400}/g)].map((m) => m[0]);
-  const sendBlob = sends.join("\n");
-  const isPluginSend = /source\s*:\s*["']adas-plugin["']/.test(sendBlob);
-
-  if (isPluginSend) {
-    if (/type\s*:\s*["']tool\.call["']/.test(sendBlob)) {
-      problems.push('sends message.type:"tool.call" — the host has NEVER accepted it, at any version (silent timeout, no error). Send message:{action:"mcp-call",payload:{requestId,connectorId,tool,args}}.');
-    } else if (!/action\s*:\s*["']mcp-call["']/.test(sendBlob)) {
-      // type:"mcp-call" is the near-miss worth naming separately: right name,
-      // wrong key. The host matches on message.action and ignores message.type.
-      if (/type\s*:\s*["']mcp-call["']/.test(sendBlob)) {
-        problems.push('sends message.TYPE:"mcp-call" — the host matches on message.ACTION for sends, so this is ignored. Use message:{action:"mcp-call",payload:{...}}.');
+  const seen = new Set();
+  const add = (p) => { if (!seen.has(p)) { seen.add(p); problems.push(p); } };
+
+  // Judge each host-directed send INDEPENDENTLY. A page may legitimately
+  // postMessage to other targets; only objects that identify themselves as the
+  // ADAS plugin channel are the protocol.
+  for (const obj of _postMessageObjects(html)) {
+    if (!/source\s*:\s*["']adas-plugin["']/.test(obj)) continue;
+
+    if (/type\s*:\s*["']tool\.call["']/.test(obj)) {
+      add('sends message.type:"tool.call" — the host has NEVER accepted it, at any version (silent timeout, no error). Send message:{action:"mcp-call",payload:{requestId,connectorId,tool,args}}.');
+    } else if (!/action\s*:\s*["']mcp-call["']/.test(obj)) {
+      if (/type\s*:\s*["']mcp-call["']/.test(obj)) {
+        add('sends message.TYPE:"mcp-call" — the host matches on message.ACTION for sends, so this is ignored. Use message:{action:"mcp-call",payload:{...}}.');
       } else {
-        problems.push('never sends action:"mcp-call" — the host ignores the message entirely.');
+        add('never sends action:"mcp-call" — the host ignores the message entirely.');
       }
     }
-    // correlationId only counts as a defect where it carries the request id.
-    if (/payload\s*:\s*\{[^}]*correlationId/.test(sendBlob) || /correlationId\s*:\s*correlationId/.test(sendBlob)) {
-      problems.push('sends the request id as correlationId — the host echoes payload.requestId, so responses never match the pending request and every call times out even when the host answered.');
+    if (/payload\s*:\s*\{[^}]*correlationId/.test(obj) || /correlationId\s*:\s*correlationId/.test(obj)) {
+      add('sends the request id as correlationId — the host echoes payload.requestId, so responses never match the pending request and every call times out even when the host answered.');
     }
   }
 
-  // RECEIVE side: only flag reading correlationId OFF THE HOST PAYLOAD.
+  // RECEIVE side: only reading correlationId OFF THE HOST PAYLOAD counts.
   if (/payload\s*(\?\.|\.)\s*correlationId/.test(html) ||
       /payload\s*&&\s*[\w$.]*payload\.correlationId/.test(html) ||
       /\{\s*correlationId[^}]*\}\s*=\s*[\w$.]*payload/.test(html)) {
-    problems.push('reads payload.correlationId from the host response — the host sends payload.requestId, so the pending request is never matched.');
+    add('reads payload.correlationId from the host response — the host sends payload.requestId, so the pending request is never matched.');
   }
   if (/type\s*===?\s*["']tool\.response["']/.test(html)) {
-    problems.push('listens for message.type:"tool.response" — the host replies with "mcp-result".');
+    add('listens for message.type:"tool.response" — the host replies with "mcp-result".');
   }
 
   return problems;
@@ -2213,6 +2241,24 @@ export const tools = [
       required: ["solution_id"],
     },
   },
+  {
+    name: "ateam_github_reconcile",
+    core: true,
+    description:
+      "JOIN a diverged dev and main. Use when ateam_github_promote returns PROMOTE_NEEDS_HUMAN or PROMOTE_PRECONDITION_FAILED — i.e. the automatic main→dev back-merge could not resolve itself.\n\n" +
+      "TRY sync_from_main FIRST; this tool calls it internally (a plain merge keeps both sides with no judgement call) and only escalates when git genuinely conflicts.\n\n" +
+      "On conflict it writes a TWO-PARENT merge commit so the histories actually join. That matters: copying one branch's tree onto the other makes the contents match while leaving NO merge base, so the very next promote conflicts again — equal content is not a reconciled history.\n\n" +
+      "Conflicting files are resolved per file, NEWEST WINS, and every decision is reported. Recency is a heuristic, not intent: read the decisions. On a real tenant main held the newer solution.json while dev held the newer widget, so a blanket choice would have reverted one of them.\n\n" +
+      "WHO NEEDS THIS: any tenant whose deploys predate the dev-routing fix carries main-only commits the platform itself wrote, and hits this on its first promote afterwards. Pass dry_run:true to see the decisions before writing anything.",
+    inputSchema: {
+      type: "object",
+      properties: {
+        solution_id: { type: "string", description: "The solution ID" },
+        dry_run: { type: "boolean", description: "Report the per-file decisions without writing the merge commit." },
+      },
+      required: ["solution_id"],
+    },
+  },
   {
     name: "ateam_github_sync_from_main",
     core: true,
@@ -5384,6 +5430,11 @@ const handlers = {
   ateam_github_promote: async ({ solution_id, label, dry_run, skip_tag }, sid) =>
     post(`/deploy/solutions/${solution_id}/promote`, { label, dry_run, skip_tag }, sid),
 
+  ateam_github_reconcile: async ({ solution_id, dry_run }, sid) => {
+    if (!solution_id) throw new Error("solution_id required");
+    return await post(`/deploy/solutions/${solution_id}/reconcile`, { dry_run: dry_run === true }, sid);
+  },
+
   ateam_github_sync_from_main: async ({ solution_id, dry_run }, sid) =>
     post(`/deploy/solutions/${solution_id}/sync-from-main`, { dry_run }, sid),