social-autoposter 1.6.22 → 1.6.24

package/bin/cli.js

@@ -96,6 +96,45 @@ function findUvBin() {
   return found && fs.existsSync(found) ? found : '';
 }
 
+// Locate the Python the MCP server is actually configured to run (SAPS_PYTHON).
+// mcp/install.mjs picks /opt/homebrew/bin/python3 (or /usr/local/bin/python3)
+// and stamps it into the MCP config, so Python deps MUST be installed into that
+// SAME interpreter. Bare `pip3`/`python3` on macOS usually resolves to the
+// Xcode CLT system python (3.9.x with pip 21.x), which is both the wrong target
+// and too old to understand --break-system-packages. Falls back to `python3`.
+function findPythonBin() {
+  const candidates = ['/opt/homebrew/bin/python3', '/usr/local/bin/python3'];
+  for (const c of candidates) {
+    if (fs.existsSync(c)) return c;
+  }
+  const which = spawnSync('command', ['-v', 'python3'], { shell: true, encoding: 'utf8' });
+  const found = (which.stdout || '').trim().split('\n')[0];
+  return found && fs.existsSync(found) ? found : 'python3';
+}
+
+// True if `<pythonBin> -m pip` is new enough (pip >= 23.0) to accept
+// --break-system-packages. Older pips treat the flag as an unknown option and
+// hard-fail, so we must not pass it blindly on the retry.
+function pipSupportsBreakSystemPackages(pythonBin) {
+  const v = spawnSync(pythonBin, ['-m', 'pip', '--version'], { encoding: 'utf8' });
+  const m = (v.stdout || '').match(/pip\s+(\d+)\.(\d+)/);
+  if (!m) return false;
+  return parseInt(m[1], 10) >= 23;
+}
+
+// Install Python packages into a specific interpreter via `<py> -m pip install`.
+// Retries with --break-system-packages only when the resolved pip supports it
+// (PEP 668 environments: Homebrew python, Debian/Ubuntu 23+). Returns the
+// spawnSync result of the last attempt.
+function pipInstall(pythonBin, args) {
+  const base = ['-m', 'pip', 'install', ...args];
+  let r = spawnSync(pythonBin, base, { stdio: 'inherit' });
+  if (r.status !== 0 && pipSupportsBreakSystemPackages(pythonBin)) {
+    r = spawnSync(pythonBin, [...base, '--break-system-packages'], { stdio: 'inherit' });
+  }
+  return r;
+}
+
 function installBrowserAgentConfigs() {
   const nodeBin = path.dirname(process.execPath);
   const uvBin = findUvBin() || path.join(HOME, '.local', 'bin', 'uv');
@@ -489,11 +528,13 @@ function installBrowserHarness() {
 
   // Step 4: ensure mcp Python package available (server.py uses `from mcp.server.fastmcp ...`).
   // server.py is shebanged through `uv run --with mcp ...` so this is belt-and-suspenders;
-  // we install it into the system Python too so a plain `python3 server.py` also works.
-  console.log('    ensuring mcp>=1.0.0 Python package is importable...');
-  let pip = spawnSync('pip3', ['install', '-q', 'mcp>=1.0.0'], { stdio: 'inherit' });
+  // we install it into the SAPS_PYTHON interpreter (the same Homebrew python the MCP
+  // server is configured to use), NOT bare pip3 which targets the Xcode CLT system python.
+  const harnessPython = findPythonBin();
+  console.log(`    ensuring mcp>=1.0.0 Python package is importable (${harnessPython})...`);
+  const pip = pipInstall(harnessPython, ['-q', 'mcp>=1.0.0']);
   if (pip.status !== 0) {
-    pip = spawnSync('pip3', ['install', '-q', 'mcp>=1.0.0', '--break-system-packages'], { stdio: 'inherit' });
+    console.warn('    WARNING: could not install mcp Python package; server.py still runs via `uv run --with mcp`.');
   }
 
   // Step 5: copy our shipped server.py into the canonical install location.
@@ -836,30 +877,29 @@ function update() {
 // browser binary; we run `playwright install chromium` after the pip install.
 function installPythonDeps() {
   const reqPath = path.join(PKG_ROOT, 'requirements.txt');
-  const base = fs.existsSync(reqPath)
-    ? ['install', '-r', reqPath, '-q']
-    : ['install', '-q', 'psycopg2-binary', 'playwright'];
-  console.log('  installing Python deps (psycopg2-binary, playwright, ...)');
-  // Debian/Ubuntu 23+ ship a PEP 668 marker that blocks pip3 against the
-  // system Python without --break-system-packages. Try without first
-  // (safer on macOS) and retry with the flag if the marker fires.
-  let r = spawnSync('pip3', base, { stdio: 'inherit' });
-  if (r.status !== 0) {
-    console.log('  retrying with --break-system-packages (PEP 668 environments)');
-    r = spawnSync('pip3', [...base, '--break-system-packages'], { stdio: 'inherit' });
-  }
+  const args = fs.existsSync(reqPath)
+    ? ['-r', reqPath, '-q']
+    : ['-q', 'psycopg2-binary', 'playwright'];
+  // Install into the SAME interpreter the MCP server runs (SAPS_PYTHON =
+  // Homebrew python), NOT bare pip3 which on macOS targets the Xcode CLT system
+  // python — deps installed there are invisible to the scripts at runtime.
+  // pipInstall() also gates --break-system-packages on pip>=23 so it doesn't
+  // hard-fail against the ancient system pip.
+  const pythonBin = findPythonBin();
+  console.log(`  installing Python deps (psycopg2-binary, playwright, ...) into ${pythonBin}`);
+  const r = pipInstall(pythonBin, args);
   if (r.status !== 0) {
-    console.warn('  WARNING: pip3 install failed — run manually:');
-    console.warn(`    pip3 ${base.join(' ')} --break-system-packages`);
+    console.warn('  WARNING: pip install failed — run manually:');
+    console.warn(`    ${pythonBin} -m pip install ${args.join(' ')} --break-system-packages`);
     return;
   }
   // Playwright needs its browser binary downloaded separately. Chromium
   // is the only engine the repo uses today; skip Firefox/WebKit.
   console.log('  installing Playwright Chromium binary (one-time, ~150MB)...');
-  const pw = spawnSync('python3', ['-m', 'playwright', 'install', 'chromium'], { stdio: 'inherit' });
+  const pw = spawnSync(pythonBin, ['-m', 'playwright', 'install', 'chromium'], { stdio: 'inherit' });
   if (pw.status !== 0) {
     console.warn('  WARNING: playwright install chromium failed — run manually:');
-    console.warn('    python3 -m playwright install chromium');
+    console.warn(`    ${pythonBin} -m playwright install chromium`);
   }
 }
 

package/mcp/dist/index.js

@@ -242,7 +242,12 @@ async function postApproved(batchId, plan) {
         return { attempted: 0, exit_code: 0, summary: "nothing approved" };
     const approvedBatch = `${batchId}_approved`;
     writePlan(approvedBatch, { ...plan, candidates: approved });
-    const res = await runPython("scripts/twitter_post_plan.py", ["--plan", planPath(approvedBatch)], { timeoutMs: 900_000 });
+    // SAPS_SKIP_CAMPAIGN_SUFFIX=1: manual/reviewed posts from this MCP draft_cycle
+    // never get the active-campaign suffix (e.g. " written with ai") appended.
+    // twitter_browser.py's reply handler reads this env (inherited through
+    // twitter_post_plan.py's subprocess). The cron pipeline doesn't set it, so the
+    // A/B disclosure experiment keeps running on autopilot/cron and on Reddit.
+    const res = await runPython("scripts/twitter_post_plan.py", ["--plan", planPath(approvedBatch)], { timeoutMs: 900_000, env: { SAPS_SKIP_CAMPAIGN_SUFFIX: "1" } });
     let summary = res.stdout.trim();
     try {
         const lines = res.stdout.trim().split("\n");

package/mcp/install.mjs

@@ -153,32 +153,13 @@ if (ok === targets.length) {
     console.log("server: " + distEntry);
     console.log("");
     console.log(
-      "AGENT INSTRUCTION: The MCP server is registered but is NOT loaded in this",
-    );
-    console.log(
-      "session yet (MCP servers load at launch, not per-tab). End your reply to the",
-    );
-    console.log(
-      "user with EXACTLY the following three steps as the FINAL lines of your response,",
-    );
-    console.log(
-      "each on its own line, in bold, numbered, with nothing after them:",
+      "The MCP server is registered but not yet loaded (MCP servers load at launch,",
     );
+    console.log("not per-tab). To finish setup:");
     console.log("");
-    console.log(
-      "  **1. Fully quit Claude now: press Cmd+Q (closing the window is not enough).**",
-    );
-    console.log("  **2. Reopen Claude.**");
-    console.log(
-      '  **3. In the message box, type:  Set me up on social-autoposter.  and send it.**',
-    );
-    console.log("");
-    console.log(
-      "Do not paraphrase those three lines, do not add commentary after them, and do",
-    );
-    console.log(
-      "not skip the bold. They are the only way the user can start the autoposter.",
-    );
+    console.log("  1. Fully quit Claude (Cmd+Q; closing the window is not enough).");
+    console.log("  2. Reopen Claude.");
+    console.log('  3. Send: "Set me up on social-autoposter."');
   }
 } else {
   console.error(

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "social-autoposter",
-  "version": "1.6.22",
+  "version": "1.6.24",
   "description": "Automated social posting pipeline for Reddit, X/Twitter, LinkedIn, and Moltbook. Install as a Claude Code agent skill.",
   "bin": {
     "social-autoposter": "bin/cli.js"

package/requirements.txt

@@ -1,7 +1,8 @@
 # social-autoposter Python deps.
-# Installed by bin/cli.js via `pip3 install -r requirements.txt` on init/update.
-# Playwright browser binaries are downloaded separately (CLI runs
-# `python3 -m playwright install chromium`).
+# Installed by bin/cli.js on init/update into the SAPS_PYTHON interpreter
+# (Homebrew python), via `<saps_python> -m pip install -r requirements.txt`.
+# Playwright browser binaries are downloaded separately (same interpreter:
+# `<saps_python> -m playwright install chromium`).
 
 psycopg2-binary
 playwright

package/scripts/ingest_human_seo_replies.py

@@ -14,9 +14,10 @@ Flow:
      `[SEO #<id>]` in the subject by default.
   3. This script polls i@m13v.com for `is:unread subject:"Re: [SEO #"`.
      For each match it: extracts the escalation id, strips quoted history
-     from the body, and UPDATEs seo_escalations SET status='replied',
-     human_reply=..., replied_at=NOW(), gmail_inbound_id=... WHERE
-     id=N AND status='pending'.
+     from the body, and PATCHes /api/v1/seo/escalations
+     (action=ingest_reply) which sets status='replied', human_reply,
+     replied_at=NOW(), gmail_inbound_id WHERE id=N AND status='pending'.
+     All state goes through the s4l.ai HTTP API (HTTP-only, no DATABASE_URL).
   4. Marks the Gmail message as read so it is not re-ingested.
   5. seo/resume_escalations.py (run from cron_seo.sh) picks up rows with
      status='replied' and re-invokes generate_page.py --resume-escalation N.
@@ -37,8 +38,10 @@ from pathlib import Path
 
 SCRIPT_DIR = Path(__file__).resolve().parent
 SEO_DIR = SCRIPT_DIR.parent / "seo"
-sys.path.insert(0, str(SEO_DIR))
-import db_helpers
+sys.path.insert(0, str(SCRIPT_DIR))
+from http_api import api_get, api_patch, load_env  # noqa: E402
+
+load_env()
 
 from google.auth.transport.requests import Request
 from google.oauth2.credentials import Credentials
@@ -175,9 +178,6 @@ def main():
         print("No candidate Gmail messages for SEO escalation replies.")
         return
 
-    conn = db_helpers.get_conn()
-    cur = conn.cursor()
-
     ingested = 0
     skipped = 0
     for c in candidates:
@@ -205,19 +205,20 @@ def main():
             skipped += 1
             continue
 
-        cur.execute(
-            "SELECT id, status, product, keyword, gmail_inbound_id "
-            "FROM seo_escalations WHERE id = %s",
-            (escalation_id,),
-        )
-        esc_row = cur.fetchone()
+        show = api_get("/api/v1/seo/escalations",
+                       query={"mode": "show", "id": escalation_id}, ok_on_404=True)
+        esc_row = None if show.get("_not_found") else (show.get("data") or {}).get("row")
         if not esc_row:
             print(f"  SKIP {gmail_id}: escalation #{escalation_id} not found")
             skipped += 1
             mark_read(service, gmail_id) if not args.dry_run else None
             continue
 
-        eid, status, product, keyword, existing_inbound = esc_row
+        eid = esc_row.get("id")
+        status = esc_row.get("status")
+        product = esc_row.get("product")
+        keyword = esc_row.get("keyword")
+        existing_inbound = esc_row.get("gmail_inbound_id")
 
         # Idempotency: if the row already has an inbound id (already ingested),
         # just mark Gmail as read and move on. If status is not pending, we
@@ -249,27 +250,19 @@ def main():
             continue
 
         try:
-            cur.execute(
-                """
-                UPDATE seo_escalations
-                SET status = 'replied',
-                    human_reply = %s,
-                    replied_at = NOW(),
-                    gmail_inbound_id = %s,
-                    updated_at = NOW()
-                WHERE id = %s AND status = 'pending'
-                """,
-                (reply_text, gmail_id, eid),
-            )
-            if cur.rowcount == 0:
+            presp = api_patch("/api/v1/seo/escalations", {
+                "id": eid,
+                "action": "ingest_reply",
+                "human_reply": reply_text,
+                "gmail_inbound_id": gmail_id,
+            })
+            updated = int((presp.get("data") or {}).get("updated") or 0)
+            if updated == 0:
                 print(f"  ERROR {gmail_id}: UPDATE matched 0 rows (race?); leaving message unread")
                 skipped += 1
-                conn.rollback()
                 continue
-            conn.commit()
         except Exception as e:
             print(f"  ERROR {gmail_id}: update failed: {e}")
-            conn.rollback()
             skipped += 1
             continue
 
@@ -280,8 +273,6 @@ def main():
         )
         ingested += 1
 
-    cur.close()
-    conn.close()
     print(f"Done. Ingested={ingested} skipped={skipped} candidates={len(candidates)}")
 
 

package/scripts/twitter_browser.py

@@ -2072,7 +2072,14 @@ def main():
                 file=sys.stderr,
             )
             sys.exit(1)
-        result = reply_to_tweet(sys.argv[2], sys.argv[3])
+        # SAPS_SKIP_CAMPAIGN_SUFFIX=1 opts this reply out of active-campaign
+        # suffixes (e.g. " written with ai"). Set ONLY by the MCP draft_cycle
+        # post path (mcp/src/index.ts::postApproved) so manual/reviewed posts
+        # land clean; the cron pipeline never sets it, so the A/B experiment
+        # keeps running there and on Reddit. Reuses the existing apply_campaigns
+        # plumbing (same flag the self-reply path uses below).
+        _skip_camp = os.environ.get("SAPS_SKIP_CAMPAIGN_SUFFIX", "").strip().lower() in ("1", "true", "yes")
+        result = reply_to_tweet(sys.argv[2], sys.argv[3], apply_campaigns=not _skip_camp)
         print(json.dumps(result, indent=2))
 
     elif cmd == "like":