@growthub/cli 0.10.0 → 0.12.0

package/assets/worker-kits/growthub-custom-workspace-starter-v1/helpers/harvest-cursor-traces.mjs

@@ -0,0 +1,288 @@
+#!/usr/bin/env node
+/**
+ * helpers/harvest-cursor-traces.mjs — Distillation Pipeline V1, Phase 1
+ *
+ * Reads Cursor agent transcript JSONL files (one folder per session, one
+ * `<uuid>.jsonl` inside it), pairs each user query with the assistant turn(s)
+ * that follow it, filters to pairs where the assistant actually executed work
+ * (≥1 `tool_use` block), and emits a single newline-delimited JSON file ready
+ * for Phase 2 (critic-grader scoring) and Phase 3 (Unsloth QLoRA export).
+ *
+ * Quality signals captured per pair (used by the grader, not graded here):
+ *   - `executedWork`     true ⇔ assistant turn produced ≥1 tool_use
+ *   - `toolUseCount`     how many tool calls were issued
+ *   - `toolNames`        deduped list of tool names invoked
+ *   - `branchesTouched`  branch names parsed out of git/gh shell commands
+ *   - `mergedToMain`     true if any branch matches a squash-merged PR on main
+ *   - `mergedPrNumbers`  PR numbers whose `headRefName` matched
+ *
+ * Squash-to-main is the highest-signal heuristic: it means the work was
+ * accepted by the maintainer and shipped. The grader can boost those rows.
+ *
+ * Usage:
+ *   node helpers/harvest-cursor-traces.mjs \
+ *     --in   /Users/antonio/.cursor/projects/Users-antonio-growthub-local/agent-transcripts \
+ *     --out  ./antonio/distillation/raw-pairs.jsonl \
+ *     --repo /Users/antonio/growthub-local            # optional: enables gh PR enrichment
+ *     --pr-limit 200                                  # optional: how many merged PRs to fetch
+ *     --min-prompt-chars 12                           # optional: drop trivial prompts
+ *
+ * No network calls beyond `gh pr list` (only when --repo is provided).
+ */
+
+import { spawnSync } from "node:child_process";
+import fs from "node:fs";
+import path from "node:path";
+
+// ---------- arg parsing ----------
+function parseArgs(argv) {
+  const a = { in: "", out: "", repo: "", prLimit: 200, minPromptChars: 12 };
+  for (let i = 0; i < argv.length; i += 1) {
+    const t = argv[i];
+    const next = () => String(argv[++i] || "").trim();
+    if (t === "--in") a.in = next();
+    else if (t === "--out") a.out = next();
+    else if (t === "--repo") a.repo = next();
+    else if (t === "--pr-limit") a.prLimit = Number(next()) || 200;
+    else if (t === "--min-prompt-chars") a.minPromptChars = Number(next()) || 0;
+    else if (t === "--help" || t === "-h") {
+      process.stdout.write(
+        "Usage: harvest-cursor-traces.mjs --in <transcripts-dir> --out <raw-pairs.jsonl> [--repo <git-repo>] [--pr-limit N] [--min-prompt-chars N]\n",
+      );
+      process.exit(0);
+    }
+  }
+  if (!a.in || !a.out) {
+    process.stderr.write("error: --in and --out are required\n");
+    process.exit(2);
+  }
+  return a;
+}
+
+const args = parseArgs(process.argv.slice(2));
+const transcriptsDir = path.resolve(args.in);
+const outPath = path.resolve(args.out);
+const repoDir = args.repo ? path.resolve(args.repo) : "";
+
+// ---------- gh squash-merge index ----------
+/** Map<branchName, { number, mergeSha, mergedAt, title }> */
+const mergedByBranch = new Map();
+if (repoDir) {
+  if (!fs.existsSync(path.join(repoDir, ".git"))) {
+    process.stderr.write(`warn: --repo ${repoDir} has no .git dir; skipping gh enrichment\n`);
+  } else {
+    const r = spawnSync(
+      "gh",
+      [
+        "pr",
+        "list",
+        "--state",
+        "merged",
+        "--limit",
+        String(args.prLimit),
+        "--json",
+        "number,title,headRefName,mergeCommit,mergedAt",
+      ],
+      { cwd: repoDir, encoding: "utf8" },
+    );
+    if (r.status === 0) {
+      try {
+        const list = JSON.parse(r.stdout || "[]");
+        for (const pr of list) {
+          if (typeof pr.headRefName === "string" && pr.headRefName.trim()) {
+            mergedByBranch.set(pr.headRefName.trim(), {
+              number: pr.number,
+              mergeSha: pr.mergeCommit?.oid || null,
+              mergedAt: pr.mergedAt || null,
+              title: pr.title || "",
+            });
+          }
+        }
+      } catch (e) {
+        process.stderr.write(`warn: gh JSON parse failed: ${e.message}\n`);
+      }
+    } else {
+      process.stderr.write(`warn: gh pr list failed (${r.status}); continuing without merge index\n`);
+    }
+  }
+}
+
+// ---------- transcript walker ----------
+function listTranscriptFiles(rootDir) {
+  if (!fs.existsSync(rootDir)) {
+    process.stderr.write(`error: transcripts dir not found: ${rootDir}\n`);
+    process.exit(2);
+  }
+  const out = [];
+  for (const entry of fs.readdirSync(rootDir, { withFileTypes: true })) {
+    const sessionDir = path.join(rootDir, entry.name);
+    if (!entry.isDirectory()) continue;
+    for (const child of fs.readdirSync(sessionDir, { withFileTypes: true })) {
+      if (child.isFile() && child.name.endsWith(".jsonl")) {
+        out.push({ sessionId: entry.name, file: path.join(sessionDir, child.name) });
+      }
+    }
+  }
+  return out.sort((a, b) => a.sessionId.localeCompare(b.sessionId));
+}
+
+// ---------- helpers ----------
+function flattenAssistantText(blocks) {
+  if (!Array.isArray(blocks)) return "";
+  return blocks
+    .filter((b) => b && b.type === "text" && typeof b.text === "string")
+    .map((b) => b.text)
+    .join("\n\n")
+    .trim();
+}
+
+function extractToolNames(blocks) {
+  if (!Array.isArray(blocks)) return [];
+  return blocks
+    .filter((b) => b && b.type === "tool_use" && typeof b.name === "string")
+    .map((b) => b.name);
+}
+
+function extractToolInputs(blocks) {
+  if (!Array.isArray(blocks)) return [];
+  return blocks
+    .filter((b) => b && b.type === "tool_use")
+    .map((b) => ({ name: b.name, input: b.input ?? {} }));
+}
+
+const BRANCH_FROM_GH_PR = /(?:gh\s+pr\s+(?:create|merge|view|checkout)[^\n]*?)(?:--head|head\s*[:=])\s+([\w./-]+)/g;
+const BRANCH_FROM_GIT_PUSH = /git\s+push\s+(?:-u\s+)?(?:origin)\s+([\w./-]+)/g;
+const BRANCH_FROM_GIT_CHECKOUT = /git\s+checkout\s+(?:-b\s+)?([\w./-]+)/g;
+const BRANCH_FROM_BRANCH_FLAG = /(?:^|\s)(?:--branch|--head|--base)\s+([\w./-]+)/g;
+
+function extractBranchesFromShell(toolUses) {
+  const out = new Set();
+  for (const tu of toolUses) {
+    if (tu.name !== "Shell") continue;
+    const cmd = String(tu.input?.command || "");
+    if (!cmd) continue;
+    for (const re of [BRANCH_FROM_GH_PR, BRANCH_FROM_GIT_PUSH, BRANCH_FROM_GIT_CHECKOUT, BRANCH_FROM_BRANCH_FLAG]) {
+      let m;
+      re.lastIndex = 0;
+      while ((m = re.exec(cmd))) {
+        const name = (m[1] || "").trim();
+        if (name && name !== "main" && name !== "HEAD") out.add(name);
+      }
+    }
+  }
+  return [...out];
+}
+
+// ---------- main pairing pass ----------
+const files = listTranscriptFiles(transcriptsDir);
+fs.mkdirSync(path.dirname(outPath), { recursive: true });
+const outStream = fs.createWriteStream(outPath, { encoding: "utf8" });
+
+const stats = {
+  sessions: 0,
+  pairsConsidered: 0,
+  pairsKept: 0,
+  pairsDroppedNoTool: 0,
+  pairsDroppedShortPrompt: 0,
+  pairsMergedToMain: 0,
+  toolNameTotals: {},
+};
+
+for (const { sessionId, file } of files) {
+  stats.sessions += 1;
+  const lines = fs.readFileSync(file, "utf8").split("\n").filter(Boolean);
+
+  /** @type {Array<{role:string,content:any}>} */
+  const turns = [];
+  for (const ln of lines) {
+    try {
+      const j = JSON.parse(ln);
+      if (j && typeof j.role === "string" && j.message?.content) {
+        turns.push({ role: j.role, content: j.message.content });
+      }
+    } catch {
+      // skip malformed line
+    }
+  }
+
+  let pairIndex = 0;
+  for (let i = 0; i < turns.length; i += 1) {
+    if (turns[i].role !== "user") continue;
+    const userBlocks = Array.isArray(turns[i].content) ? turns[i].content : [];
+    const userText = userBlocks
+      .filter((b) => b && b.type === "text" && typeof b.text === "string")
+      .map((b) => b.text)
+      .join("\n\n")
+      .trim();
+    if (!userText) continue;
+
+    // Collect every consecutive assistant turn until the next user turn.
+    const assistantBlocks = [];
+    let j = i + 1;
+    while (j < turns.length && turns[j].role !== "user") {
+      const c = turns[j].content;
+      if (Array.isArray(c)) assistantBlocks.push(...c);
+      j += 1;
+    }
+    if (assistantBlocks.length === 0) continue;
+
+    stats.pairsConsidered += 1;
+
+    if (userText.length < args.minPromptChars) {
+      stats.pairsDroppedShortPrompt += 1;
+      continue;
+    }
+
+    const toolUses = extractToolInputs(assistantBlocks);
+    if (toolUses.length === 0) {
+      stats.pairsDroppedNoTool += 1;
+      continue;
+    }
+
+    const assistantText = flattenAssistantText(assistantBlocks);
+    const toolNames = [...new Set(extractToolNames(assistantBlocks))];
+    for (const n of toolNames) stats.toolNameTotals[n] = (stats.toolNameTotals[n] || 0) + 1;
+
+    const branchesTouched = extractBranchesFromShell(toolUses);
+    const mergedPrNumbers = [];
+    for (const b of branchesTouched) {
+      const hit = mergedByBranch.get(b);
+      if (hit) mergedPrNumbers.push(hit.number);
+    }
+    const mergedToMain = mergedPrNumbers.length > 0;
+    if (mergedToMain) stats.pairsMergedToMain += 1;
+
+    const row = {
+      sessionId,
+      pairIndex,
+      inputPrompt: userText,
+      agentOutput: assistantText,
+      executedWork: true,
+      toolUseCount: toolUses.length,
+      toolNames,
+      branchesTouched,
+      mergedToMain,
+      mergedPrNumbers,
+    };
+    outStream.write(`${JSON.stringify(row)}\n`);
+    stats.pairsKept += 1;
+    pairIndex += 1;
+  }
+}
+
+await new Promise((resolve) => outStream.end(resolve));
+
+// ---------- summary ----------
+process.stdout.write(
+  JSON.stringify(
+    {
+      ok: true,
+      out: outPath,
+      transcriptsDir,
+      mergeIndexEntries: mergedByBranch.size,
+      ...stats,
+    },
+    null,
+    2,
+  ) + "\n",
+);

package/assets/worker-kits/growthub-custom-workspace-starter-v1/helpers/upload-graded-traces.mjs

@@ -0,0 +1,128 @@
+#!/usr/bin/env node
+/**
+ * helpers/upload-graded-traces.mjs — Distillation Pipeline V1, Phase 2.5
+ *
+ * Reads graded pairs JSONL (Phase 2 output) and appends rows whose
+ * qualityScore >= --min-score into the live workspace's `training-traces`
+ * dataModel object via PATCH /api/workspace.
+ *
+ * - Append-only. Existing rows are preserved.
+ * - Marks every uploaded row with `exported: "false"` so the export script
+ *   can pick them up later.
+ * - Truncates `agentOutput` to keep the workspace config file lean.
+ *
+ * Usage:
+ *   node helpers/upload-graded-traces.mjs \
+ *     --in        ./antonio/distillation/graded-batch-001.jsonl \
+ *     --workspace http://localhost:3000 \
+ *     --traces-object training-traces \
+ *     --min-score 4 \
+ *     --max-output-chars 4000
+ */
+
+import fs from "node:fs";
+import path from "node:path";
+
+function parseArgs(argv) {
+  const a = {
+    in: "",
+    workspace: "http://localhost:3000",
+    tracesObject: "training-traces",
+    minScore: 4,
+    maxOutputChars: 4000,
+  };
+  for (let i = 0; i < argv.length; i += 1) {
+    const t = argv[i];
+    const next = () => String(argv[++i] || "").trim();
+    if (t === "--in") a.in = next();
+    else if (t === "--workspace") a.workspace = next().replace(/\/+$/, "");
+    else if (t === "--traces-object") a.tracesObject = next();
+    else if (t === "--min-score") a.minScore = Number(next()) || 4;
+    else if (t === "--max-output-chars") a.maxOutputChars = Number(next()) || 4000;
+    else if (t === "--help" || t === "-h") {
+      process.stdout.write(
+        "Usage: upload-graded-traces.mjs --in <graded.jsonl> [--workspace URL] [--traces-object id] [--min-score N] [--max-output-chars N]\n",
+      );
+      process.exit(0);
+    }
+  }
+  if (!a.in) {
+    process.stderr.write("error: --in is required\n");
+    process.exit(2);
+  }
+  return a;
+}
+
+const args = parseArgs(process.argv.slice(2));
+const inAbs = path.resolve(args.in);
+if (!fs.existsSync(inAbs)) {
+  process.stderr.write(`error: input not found: ${inAbs}\n`);
+  process.exit(2);
+}
+
+async function getObjects() {
+  const r = await fetch(`${args.workspace}/api/workspace`, { cache: "no-store" });
+  if (!r.ok) throw new Error(`GET /api/workspace ${r.status}`);
+  return (await r.json()).workspaceConfig.dataModel.objects;
+}
+async function patchObjects(objects) {
+  const r = await fetch(`${args.workspace}/api/workspace`, {
+    method: "PATCH",
+    headers: { "content-type": "application/json" },
+    body: JSON.stringify({ dataModel: { objects } }),
+  });
+  if (!r.ok) throw new Error(`PATCH ${r.status}: ${(await r.text()).slice(0, 300)}`);
+}
+
+const lines = fs.readFileSync(inAbs, "utf8").split("\n").filter(Boolean);
+const candidates = [];
+for (const ln of lines) {
+  try {
+    const j = JSON.parse(ln);
+    if (Number(j.qualityScore) >= args.minScore) candidates.push(j);
+  } catch {
+    /* skip */
+  }
+}
+
+if (candidates.length === 0) {
+  process.stdout.write(JSON.stringify({ ok: true, uploaded: 0, reason: "no rows >= --min-score" }) + "\n");
+  process.exit(0);
+}
+
+const objects = await getObjects();
+const tracesIdx = objects.findIndex((o) => o.id === args.tracesObject);
+if (tracesIdx < 0) {
+  process.stderr.write(`error: object ${args.tracesObject} not found in workspace\n`);
+  process.exit(3);
+}
+const tracesObj = objects[tracesIdx];
+const newRows = candidates.map((p) => ({
+  sessionDate: p.gradedAt || new Date().toISOString(),
+  inputPrompt: String(p.inputPrompt || "").slice(0, args.maxOutputChars),
+  agentOutput: String(p.agentOutput || "").slice(0, args.maxOutputChars),
+  qualityScore: String(p.qualityScore),
+  reason: String(p.qualityReason || ""),
+  exported: "false",
+}));
+
+const nextObjects = objects.map((o, i) =>
+  i !== tracesIdx ? o : { ...o, rows: [...(o.rows || []), ...newRows] },
+);
+await patchObjects(nextObjects);
+
+const after = await getObjects();
+const finalCount = after.find((o) => o.id === args.tracesObject)?.rows?.length || 0;
+process.stdout.write(
+  JSON.stringify(
+    {
+      ok: true,
+      candidates: candidates.length,
+      uploaded: newRows.length,
+      tracesTotalRows: finalCount,
+      sourceFile: path.basename(inAbs),
+    },
+    null,
+    2,
+  ) + "\n",
+);

package/assets/worker-kits/growthub-custom-workspace-starter-v1/kit.json

@@ -43,10 +43,15 @@
     "templates/deployment-plan.md",
     "templates/project.md",
     "templates/self-eval.md",
+    "templates/seeded-configs/alignment-loop.config.json",
     "helpers/README.md",
     "helpers/propose-capability.mjs",
     "helpers/promote-capability.mjs",
     "helpers/check-self-improving-health.sh",
+    "helpers/harvest-cursor-traces.mjs",
+    "helpers/grade-raw-pairs.mjs",
+    "helpers/upload-graded-traces.mjs",
+    "helpers/export-training-traces.mjs",
     "skills/README.md",
     "examples/workspace-sample.md",
     "docs/starter-kit-overview.md",
@@ -98,9 +103,18 @@
     "apps/workspace/lib/adapters/sandboxes/adapter-loader.js",
     "apps/workspace/lib/adapters/sandboxes/adapters/README.md",
     "apps/workspace/lib/adapters/sandboxes/default-local-agent-host.js",
+    "apps/workspace/lib/adapters/sandboxes/default-local-intelligence.js",
     "apps/workspace/lib/adapters/sandboxes/default-local-process.js",
     "apps/workspace/lib/adapters/sandboxes/index.js",
     "apps/workspace/lib/adapters/sandboxes/sandbox-adapter-registry.js",
+    "apps/workspace/lib/workspace-helper.js",
+    "apps/workspace/lib/workspace-helper-apply.js",
+    "apps/workspace/app/api/workspace/helper/query/route.js",
+    "apps/workspace/app/api/workspace/helper/apply/route.js",
+    "apps/workspace/app/api/workspace/helper/receipts/route.js",
+    "apps/workspace/app/data-model/page.jsx",
+    "apps/workspace/app/data-model/components/DataModelShell.jsx",
+    "apps/workspace/app/data-model/components/HelperSidecar.jsx",
     "apps/workspace/lib/adapters/payments/index.js",
     "apps/workspace/lib/adapters/persistence/index.js",
     "apps/workspace/lib/adapters/persistence/postgres.js",
@@ -139,11 +153,16 @@
       "templates/self-eval.md",
       "templates/deployment-handoff.md",
       "templates/supabase-setup-plan.md",
+      "templates/seeded-configs/alignment-loop.config.json",
       "helpers",
       "helpers/README.md",
       "helpers/propose-capability.mjs",
       "helpers/promote-capability.mjs",
       "helpers/check-self-improving-health.sh",
+      "helpers/harvest-cursor-traces.mjs",
+      "helpers/grade-raw-pairs.mjs",
+      "helpers/upload-graded-traces.mjs",
+      "helpers/export-training-traces.mjs",
       "skills",
       "skills/README.md",
       "docs",