Expose recovery job state through the API surface

.claude/scripts/install-claude.sh

@@ -9,8 +9,8 @@ if [ "${CI:-}" != "true" ]; then
   exit 1
 fi
 
-VERSION="2.1.85"
-CHECKSUM="ff0b23dba11c97a53386c61ebe47d46d768a8ad33f98c7d22186c9a63f179f4d"
+VERSION="2.1.98"
+CHECKSUM="d40827b5aa8d737a7eb68e3aad990b80e2521540a6bc8a405259b63b25d42ed8"
 GCS_BUCKET="https://storage.googleapis.com/claude-code-dist-86c565f3-f756-42ad-8dfa-d59b1c096819/claude-code-releases"
 PLATFORM="linux-x64"
 

.claude/settings.json

@@ -2,7 +2,7 @@
   "$schema": "https://json.schemastore.org/claude-code-settings.json",
   "statusLine": {
     "type": "command",
-    "command": "$CLAUDE_PROJECT_DIR/.claude/statusline.sh"
+    "command": "$CLAUDE_PROJECT_DIR/.claude/scripts/statusline.sh"
   },
   "hooks": {
     "PreToolUse": [

.claude/skills/pr-review/SKILL.md

@@ -74,8 +74,17 @@ For each issue found, use the `add-review-comment` skill to post review comments
 
 ### 6. Approve the PR
 
-Approve the PR when there are no issues or only minor issues.
+Approve the PR when there are no issues or only minor issues, but **only if the PR author has the `admin` or `maintain` role**.
+
+First, check the PR author's role:
 
 ```bash
-gh pr review <PR_NUMBER> --repo <owner/repo> --approve
+author=$(gh api repos/<owner>/<repo>/pulls/<PR_NUMBER> --jq '.user.login')
+gh api repos/<owner>/<repo>/collaborators/"$author"/permission --jq '.role_name'
 ```
+
+- If the role is `admin` or `maintain` -> approve the PR:
+  ```bash
+  gh pr review <PR_NUMBER> --repo <owner/repo> --approve
+  ```
+- Otherwise (including API errors, e.g., 404 for non-collaborators) -> do NOT approve. Do not mention the reason for not approving in the review.

.claude/skills/pyproject.toml

@@ -2,13 +2,7 @@
 name = "skills"
 version = "0.1.0"
 requires-python = ">=3.10"
-dependencies = [
-  "aiohttp",
-  "claude-agent-sdk",
-  "pydantic",
-  "tiktoken",
-  "typing_extensions",
-]
+dependencies = ["aiohttp", "claude-agent-sdk", "pydantic", "typing_extensions"]
 
 [project.scripts]
 skills = "skills.cli:main"

.claude/skills/src/skills/commands/analyze_ci.py

@@ -14,7 +14,6 @@
 from dataclasses import dataclass
 from typing import Any
 
-import tiktoken
 from claude_agent_sdk import (
     AssistantMessage,
     ClaudeAgentOptions,
@@ -26,6 +25,7 @@
 from skills.github import GitHubClient, Job, JobStep, get_github_token
 
 MAX_LOG_TOKENS = 100_000
+CHARS_PER_TOKEN = 2
 
 
 @dataclass
@@ -111,21 +111,17 @@ async def compact_logs(lines: AsyncIterator[str]) -> str:
             result.append(line)
 
     logs = "\n".join(result)
-    tokens = tiktoken.get_encoding("p50k_base").encode(logs)
-    log(f"Compacted logs: {len(tokens):,} tokens")
+    log(f"Compacted logs: {len(logs) // CHARS_PER_TOKEN:,} tokens")
     return logs
 
 
 def truncate_logs(logs: str, max_tokens: int = MAX_LOG_TOKENS) -> str:
     """Truncate logs to fit within token limit, keeping the end (where errors are)."""
-    # Note: tiktoken token count is an estimation and may differ slightly from
-    # the official token count API
-    tokenizer = tiktoken.get_encoding("p50k_base")
-    tokens = tokenizer.encode(logs)
-    if len(tokens) <= max_tokens:
+    estimated_tokens = len(logs) // CHARS_PER_TOKEN
+    if estimated_tokens <= max_tokens:
         return logs
-    log(f"Truncating logs from {len(tokens):,} to {max_tokens:,} tokens")
-    truncated = tokenizer.decode(tokens[-max_tokens:])
+    log(f"Truncating logs from {estimated_tokens:,} to {max_tokens:,} tokens")
+    truncated = logs[-(max_tokens * CHARS_PER_TOKEN) :]
     return f"(showing last {max_tokens:,} tokens)\n{truncated}"
 
 

.github/actions/check-component-ids/action.yml

@@ -0,0 +1,10 @@
+name: "check-component-ids"
+description: "Verify that all componentIds in the MLflow UI are registered in the componentId registry and vice versa."
+runs:
+  using: "composite"
+  steps:
+    - uses: actions/setup-node@53b83947a5a98c8d113130e565377fae1a50d02f # v6.3.0
+      with:
+        node-version: 20
+    - run: node ${GITHUB_ACTION_PATH}/index.js
+      shell: bash

.github/actions/check-component-ids/index.js

@@ -0,0 +1,41 @@
+const { extractComponentIdsFromSource } = require("./utils");
+
+const registry = require("./componentId-registry");
+
+// --- Main ---
+const codeIds = extractComponentIdsFromSource(__dirname);
+const registryKeys = new Set(Object.keys(registry));
+
+// Check 1: componentIds in code but not in registry
+const unregistered = [...codeIds].filter((id) => !registryKeys.has(id)).sort();
+
+// Check 2: componentIds in registry but not in code (stale)
+const stale = [...registryKeys].filter((id) => !codeIds.has(id)).sort();
+
+let failed = false;
+
+if (unregistered.length > 0) {
+  failed = true;
+  console.error(
+    `\n❌ Found ${unregistered.length} componentId(s) in code but NOT in the registry:\n`
+  );
+  for (const id of unregistered) {
+    console.error(`  + ${id}`);
+  }
+  console.error("\nAdd these to .github/actions/check-component-ids/componentId-registry.js");
+}
+
+if (stale.length > 0) {
+  failed = true;
+  console.error(`\n❌ Found ${stale.length} stale componentId(s) in registry but NOT in code:\n`);
+  for (const id of stale) {
+    console.error(`  - ${id}`);
+  }
+  console.error("\nRemove these from .github/actions/check-component-ids/componentId-registry.js");
+}
+
+if (failed) {
+  process.exit(1);
+} else {
+  console.log(`✅ componentId registry is in sync. ${registryKeys.size} entries verified.`);
+}

.github/actions/check-component-ids/regenerate.js

@@ -0,0 +1,70 @@
+#!/usr/bin/env node
+
+/**
+ * Regenerates the componentId registry from source code.
+ *
+ * Usage (from repo root):
+ *   node .github/actions/check-component-ids/regenerate.js
+ */
+
+const fs = require("fs");
+const path = require("path");
+const { extractComponentIdsFromSource } = require("./utils");
+
+const codeIds = extractComponentIdsFromSource(__dirname);
+const sorted = [...codeIds].sort();
+
+// Group by prefix for readability
+const groups = {};
+for (const id of sorted) {
+  let prefix;
+  if (id.startsWith("codegen_")) {
+    prefix = "Codegen (auto-generated)";
+  } else if (id.startsWith("mlflow.")) {
+    const parts = id.split(".");
+    prefix = parts[0] + "." + parts[1];
+  } else if (id.startsWith("shared.")) {
+    const parts = id.split(".");
+    prefix = parts[0] + "." + parts[1];
+  } else {
+    prefix = "Other";
+  }
+  if (!groups[prefix]) groups[prefix] = [];
+  groups[prefix].push(id);
+}
+
+// Load existing registry to preserve descriptions
+let existingDescriptions = {};
+try {
+  existingDescriptions = require("./componentId-registry");
+} catch {
+  // First run or broken registry — start fresh
+}
+
+let output = `/**
+ * Curated registry of all componentIds used in the MLflow UI.
+ *
+ * Every static componentId string literal in non-test source files must
+ * have an entry here. The CI job \`check-component-ids\` verifies this
+ * bidirectionally: code IDs must be in the registry, and registry
+ * entries must exist in code.
+ *
+ * Format: key = componentId string, value = optional description of the
+ * component (blank by default, especially for generated entries)
+ */
+module.exports = {\n`;
+
+for (const gk of Object.keys(groups).sort()) {
+  output += `  // -- ${gk} --\n`;
+  for (const id of groups[gk]) {
+    const escaped = id.replace(/"/g, '\\"');
+    const desc = (existingDescriptions[id] || "").replace(/"/g, '\\"');
+    output += `  "${escaped}": "${desc}",\n`;
+  }
+  output += "\n";
+}
+output += "};\n";
+
+const outPath = path.join(__dirname, "componentId-registry.js");
+fs.writeFileSync(outPath, output);
+console.log(`✅ Registry regenerated with ${sorted.length} entries at ${outPath}`);

.github/actions/check-component-ids/utils.js

@@ -0,0 +1,66 @@
+const fs = require("fs");
+const path = require("path");
+
+const EXTENSIONS = [".js", ".jsx", ".ts", ".tsx"];
+// Skip test files — they don't need registered componentIds
+const TEST_PATTERN = /\.test\.[jt]sx?$/;
+
+const EXTRACT_PATTERNS = [
+  /(?:componentId|data-component-id)=["']([^"']+)["']/g,
+  /componentId:\s*["']([^"']+)["']/g,
+  // Match static strings inside JSX expressions like componentId={"value"},
+  // componentId={cond ?? "fallback"}, componentId={cond ? "a" : "b"}, etc.
+  // Uses [^\n}]* to avoid matching across lines.
+  /componentId=\{[^\n}]*["']([^"'\n`]+)["'][^\n}]*\}/g,
+];
+
+function findFiles(dir) {
+  const results = [];
+  function walk(d) {
+    for (const entry of fs.readdirSync(d, { withFileTypes: true })) {
+      const full = path.join(d, entry.name);
+      if (entry.isDirectory() && !entry.name.startsWith(".") && entry.name !== "node_modules") {
+        walk(full);
+      } else if (
+        entry.isFile() &&
+        EXTENSIONS.some((ext) => full.endsWith(ext)) &&
+        !TEST_PATTERN.test(full)
+      ) {
+        results.push(full);
+      }
+    }
+  }
+  walk(dir);
+  return results;
+}
+
+function extractComponentIds(files) {
+  const ids = new Set();
+  for (const file of files) {
+    const content = fs.readFileSync(file, "utf8");
+    for (const pat of EXTRACT_PATTERNS) {
+      pat.lastIndex = 0;
+      let m;
+      while ((m = pat.exec(content)) !== null) {
+        ids.add(m[1]);
+      }
+    }
+  }
+  return ids;
+}
+
+/**
+ * Extract all static componentIds from the MLflow UI source directory.
+ * @param {string} actionDir - path to this action's directory (used to resolve the repo root)
+ * @returns {Set<string>} set of componentId strings found in source
+ */
+function extractComponentIdsFromSource(actionDir) {
+  const srcDir = path.resolve(
+    process.env.GITHUB_WORKSPACE || path.join(actionDir, "../../.."),
+    "mlflow/server/js/src"
+  );
+  const files = findFiles(srcDir);
+  return extractComponentIds(files);
+}
+
+module.exports = { extractComponentIdsFromSource };

.github/actions/setup-python/action.yml

@@ -8,6 +8,10 @@ inputs:
     description: "Whether to pin to a specific micro version for Anaconda compatibility. Set to false for workflows that don't need conda/pyenv to hit the runner's pre-installed Python cache and avoid a ~9s download."
     required: false
     default: "true"
+  enable-uv-cache:
+    description: "Enable uv's built-in cache. Set to false for short jobs where cache restore/save overhead isn't worth it."
+    required: false
+    default: "true"
 outputs:
   python-version:
     description: "The installed python version"
@@ -52,6 +56,7 @@ runs:
     - uses: astral-sh/setup-uv@37802adc94f370d6bfd71619e3f0bf239e1f3b78 # v7.6.0
       with:
         version: "0.10.12"
+        enable-cache: ${{ inputs.enable-uv-cache }}
     - run: |
         # The default `first-index` strategy is too strict. Use `unsafe-first-match` instead.
         # https://docs.astral.sh/uv/configuration/environment/#uv_index_strategy

.github/ui-preview/app.py

@@ -20,10 +20,12 @@ def setup():
     _logger.info("Extracting UI assets to %s", target_dir)
     subprocess.check_call(["tar", "xzf", tar_path, "-C", target_dir])
 
-    # Generate demo data
+    # Generate demo data. Always refresh so the preview app reflects the latest
+    # demo content (e.g. new trace types) even if the SQLite database persisted
+    # from a previous deploy with stale demo data.
     os.environ["MLFLOW_TRACKING_URI"] = "sqlite:///mlflow.db"
     _logger.info("Generating demo data...")
-    generate_all_demos()
+    generate_all_demos(refresh=True)
     _logger.info("Demo data generated.")
 
 

.github/workflows/require-core-maintainer-approval.js → .github/workflows/approval.js

@@ -70,7 +70,10 @@ module.exports = async ({ github, context, core }) => {
     pull_number: context.issue.number,
   });
   const maintainerApproved = reviews.some(
-    ({ state, user: { login } }) => state === "APPROVED" && maintainers.includes(login)
+    ({ state, user }) =>
+      state === "APPROVED" &&
+      (maintainers.includes(user.login) ||
+        (user.type.toLowerCase() === "bot" && user.login === "mlflow-app[bot]"))
   );
 
   const { pull_request: pr } = context.payload;

.github/workflows/maintainer-approval.yml → .github/workflows/approval.yml

@@ -1,4 +1,4 @@
-name: Maintainer approval
+name: Approval
 
 on:
   pull_request_target:
@@ -24,5 +24,5 @@ jobs:
         with:
           retries: 3
           script: |
-            const script = require(`${process.env.GITHUB_WORKSPACE}/.github/workflows/require-core-maintainer-approval.js`);
+            const script = require(`${process.env.GITHUB_WORKSPACE}/.github/workflows/approval.js`);
             await script({ context, github, core });

.github/workflows/cross-version-tests.yml

@@ -74,6 +74,9 @@ jobs:
           ref: ${{ github.event.inputs.ref }}
       - uses: ./.github/actions/untracked
       - uses: ./.github/actions/setup-python
+        with:
+          pin-micro-version: false
+          enable-uv-cache: false
       - name: Install dependencies
         run: |
           uv pip install --system -r dev/requirements.txt

.github/workflows/duplicate-prs.js

@@ -25,6 +25,7 @@ const QUERY = `
           url
           author { login }
           authorAssociation
+          labels(first: 20) { nodes { name } }
           closingIssuesReferences(first: 10) {
             nodes {
               number
@@ -42,6 +43,9 @@ const shouldProcessPR = (pr) => {
   if (memberAssociations.includes(pr.authorAssociation)) {
     return false;
   }
+  // Skip PRs already labeled as duplicate
+  const labels = pr.labels?.nodes?.map((l) => l.name) ?? [];
+  if (labels.includes(DUPLICATE_LABEL)) return false;
   return true;
 };