@m13v/s4l 1.7.6-rc.8 → 1.7.6

package/mcp/dist/setup.js

@@ -192,11 +192,19 @@ export function projectExists(name) {
         return false;
     }
 }
-// Strip stray JSON-array syntax that leaks into a single topic when a
-// stringified array was split on commas (leading "[", trailing "]", and the
-// surrounding quotes on each element). Without this, topics like '["AI video
-// generation"' and '"ex-Google"]' get seeded verbatim and poison every search
-// query with literal [, ] and " characters (Karol, 2026-06-30).
+// Strip stray JSON-array syntax that leaks into a single element ONLY when a
+// stringified array had to be split on commas (leading "[", trailing "]", and
+// the surrounding quotes on each element). Without this, topics like '["AI
+// video generation"' and '"ex-Google"]' get seeded verbatim and poison every
+// search query with literal [, ] and " characters (Karol, 2026-06-30).
+//
+// IMPORTANT: apply this ONLY on the naive-split fallback path. On the clean
+// paths (a real array, or a valid stringified array that JSON.parse handled)
+// each element is already clean, so stripping surrounding quotes there would
+// destroy legitimate phrase-match quotes in a search query — e.g.
+// '"agentic coding" min_faves:10' would lose its LEADING " and seed as
+// 'agentic coding" min_faves:10', weakening the phrase match (reported
+// 2026-07-17). Clean paths therefore only get trimmed.
 function stripTopicSyntax(x) {
     let s = String(x).trim();
     s = s.replace(/^\[+/, "").replace(/\]+$/, "").trim();
@@ -213,6 +221,9 @@ export function normalizeStringList(t) {
     if (t == null)
         return undefined;
     let raw;
+    // Elements only carry stray JSON-array syntax when we reach the naive
+    // comma/newline split below; on the clean paths they're already delimited.
+    let fromNaiveSplit = false;
     if (Array.isArray(t)) {
         raw = t;
     }
@@ -227,9 +238,18 @@ export function normalizeStringList(t) {
                 parsed = null;
             }
         }
-        raw = Array.isArray(parsed) ? parsed : s.split(/[,\n]/);
+        if (Array.isArray(parsed)) {
+            raw = parsed;
+        }
+        else {
+            raw = s.split(/[,\n]/);
+            fromNaiveSplit = true;
+        }
     }
-    return raw.map((x) => stripTopicSyntax(String(x))).filter(Boolean);
+    const clean = fromNaiveSplit
+        ? stripTopicSyntax
+        : (x) => String(x).trim();
+    return raw.map((x) => clean(String(x))).filter(Boolean);
 }
 function normalizeTopics(t) {
     return normalizeStringList(t);

package/mcp/dist/telemetry.js

@@ -139,12 +139,21 @@ export function checkVersionChange() {
     }
     const isFirstBoot = from === null;
     const changed = !isFirstBoot && from !== to;
-    try {
-        fs.mkdirSync(snapshotStateDir(), { recursive: true });
-        fs.writeFileSync(lastVersionPath(), JSON.stringify({ version: to, seen_at: new Date().toISOString() }), "utf-8");
-    }
-    catch {
-        /* best-effort; a failed write just means the next boot re-detects from "from: null" */
+    // Write ONLY when the version actually changed (or first boot / unreadable
+    // file). The old rewrite-every-boot kept the file's timestamp meaning "when
+    // the server last booted", which reads like an install time and misled a
+    // 2026-07-18 investigation into concluding an update hadn't applied. With
+    // the guard, running_since is exactly what it says: when THIS version first
+    // booted here. Same-version boots leave the file untouched, so detection
+    // semantics are unchanged (the file already holds the current version).
+    if (changed || isFirstBoot) {
+        try {
+            fs.mkdirSync(snapshotStateDir(), { recursive: true });
+            fs.writeFileSync(lastVersionPath(), JSON.stringify({ version: to, running_since: new Date().toISOString() }), "utf-8");
+        }
+        catch {
+            /* best-effort; a failed write just means the next boot re-detects from "from: null" */
+        }
     }
     if (changed) {
         console.error(`[social-autoposter-mcp] self-update detected: ${from} -> ${to}`);

package/mcp/dist/version.json

@@ -1,4 +1,4 @@
 {
-  "version": "1.7.6-rc.8",
-  "installedAt": "2026-07-17T17:51:37.267Z"
+  "version": "1.7.6",
+  "installedAt": "2026-07-18T17:32:42.014Z"
 }

package/mcp/manifest.json

@@ -2,7 +2,7 @@
   "dxt_version": "0.1",
   "name": "social-autoposter",
   "display_name": "S4L",
-  "version": "1.7.6-rc.8",
+  "version": "1.7.6",
   "description": "Draft, review, approve, and autopilot X/Twitter posts.",
   "long_description": "## **⚠️ The disclaimer above is generic Claude boilerplate.** Anthropic shows the same warning on every plugin regardless of what it does; any plugin has the same level of access as any app you download from the internet.\n\nS4L is an open source product developed by Mediar.ai Incorporated, a VC-backed San Francisco-based startup.\n\nTo get started:\n\n1\\. Copy this prompt: **Set me up on S4L plugin end to end**\n\n2\\. Quit with CMD+Q, reopen Claude, paste into a new chat.\n\nWhat happens next:\n\n* About every 5 minutes S4L scans X for posts that match your topics and drafts replies in your voice.\n* Drafts show up as review cards, usually the first within a few minutes. Nothing is posted automatically; you approve each one.\n* Posting autopilot stays off until you explicitly turn it on.",
   "author": {
@@ -33,7 +33,7 @@
       "description": "Configure or edit a project"
     },
     {
-      "name": "post_drafts",
+      "name": "approve_drafts",
       "description": "Post chosen drafts"
     },
     {
@@ -48,6 +48,14 @@
       "name": "restart_menubar",
       "description": "Restart the S4L menu bar app"
     },
+    {
+      "name": "run_drafting",
+      "description": "Run the drafting pipeline now"
+    },
+    {
+      "name": "drafting_status",
+      "description": "Live drafting-pipeline status + next-run countdown"
+    },
     {
       "name": "posting_volume",
       "description": "Read or set posting volume (Aggressive / Steady / Chill)"

package/mcp/menubar/s4l_browser_foreground.py

@@ -46,7 +46,13 @@ import s4l_log_relay
 
 # A managed harness Chrome is one launched on a profile under this marker
 # (browser-harness = twitter 9555, browser-harness-linkedin = 9556, ...).
-_PROFILE_MARKER = os.path.join(".claude", "browser-profiles", "browser-harness")
+# Any managed harness profile under browser-profiles/ counts — matching the
+# literal "browser-harness" prefix silently EXCLUDED the reddit harness
+# (profile "reddit-harness"), so reddit-Chrome activations went unlogged and
+# unattributable for days (2026-07-17). The remote-debugging-port requirement
+# in the check below keeps the user's own Chrome / MCP-agent profiles (which
+# use the debugging PIPE, not a port) out of scope.
+_PROFILE_MARKER = os.path.join(".claude", "browser-profiles", "")
 
 # Within this window, repeats of the same (cause, pid) are counted, not emitted.
 # A screencast-reconnect storm raises Chrome every few seconds; one line per

package/mcp/menubar/s4l_card.py

@@ -331,7 +331,7 @@ def _truncate(s, n=320):
     return s if len(s) <= n else s[: n - 1] + "…"
 
 
-def _set_thread_text(thread_tv, text, link_url, *, font_size=12):
+def _set_thread_text(thread_tv, text, link_url, *, font_size=12, title=None):
     """Set the thread-quote text on `thread_tv`: the FULL text, never
     trimmed (2026-07-16 user direction -- the box scrolls instead when it
     doesn't fit; see its NSScrollView wrapper in _render()). A leading
@@ -340,8 +340,15 @@ def _set_thread_text(thread_tv, text, link_url, *, font_size=12):
     of how long the thread is or where the box happens to be scrolled to
     (the old trailing-link placement could end up scrolled out of view, or
     -- before scrolling existed at all -- get silently truncated off the
-    end; see the superseded _fit_thread_body this replaced)."""
+    end; see the superseded _fit_thread_body this replaced).
+
+    `title` (2026-07-17 user direction, Reddit): when given, it's drawn BOLD
+    on its own line ahead of `text`, so a Reddit card shows the post title in
+    bold with the selftext body beneath it in the same quote box. Twitter and
+    title-only Reddit threads pass title=None and render `text` alone,
+    exactly as before."""
     text = (text or "").strip()
+    title = (title or "").strip()
     body = NSMutableAttributedString.alloc().initWithString_attributes_("", {})
     if link_url:
         body.appendAttributedString_(
@@ -356,6 +363,19 @@ def _set_thread_text(thread_tv, text, link_url, *, font_size=12):
                 },
             )
         )
+    if title:
+        # Bold title on its own line; a trailing newline separates it from the
+        # body only when there IS a body (a title with no selftext shouldn't
+        # leave a dangling blank line).
+        body.appendAttributedString_(
+            NSAttributedString.alloc().initWithString_attributes_(
+                title + ("\n" if text else ""),
+                {
+                    NSFontAttributeName: NSFont.boldSystemFontOfSize_(font_size),
+                    NSForegroundColorAttributeName: NSColor.labelColor(),
+                },
+            )
+        )
     body.appendAttributedString_(
         NSAttributedString.alloc().initWithString_attributes_(
             text,
@@ -425,8 +445,26 @@ def _engagement_line(stats):
     return " · ".join(parts)
 
 
+def _compact_arm(variant):
+    """Shrink a draft_prompt A/B variant to a heading-sized token so the arm
+    fits inline next to project/lane/viral: 'treatment_v4' -> 't v4',
+    'control_v4' -> 'c v4'. Anything that isn't a treatment/control_vN shape
+    (future arm names, retired bare 'treatment'/'control') falls back to the
+    raw variant, so no arm goes unlabelled. The full one-sentence meaning
+    still lives in the details popover via active_experiments.describe()."""
+    v = (variant or "").strip()
+    if not v:
+        return ""
+    m = re.match(r"(treatment|control)(?:_v?(\w+))?$", v)
+    if not m:
+        return v
+    arm = "t" if m.group(1) == "treatment" else "c"
+    ver = m.group(2)
+    return f"{arm} v{ver}" if ver else arm
+
+
 def _reply_heading_suffix(d):
-    """Concise 'project/lane · viral N' readout appended straight onto the
+    """Concise 'project/lane · viral N · t v4' readout appended straight onto the
     'Reply (editable):' heading (2026-07-07), so the two facts a reviewer
     checks first don't require opening the details eye. Pulled from the same
     fields _details_lines used to render as 'Project:'/'Lane:'/'Virality
@@ -465,6 +503,15 @@ def _reply_heading_suffix(d):
             bits.append(f"viral {float(v):g}")
         except (TypeError, ValueError):
             pass
+    # The drafting A/B arm (treatment vs control), compacted to 't v4'/'c v4'
+    # so the reviewer sees which prompt variant produced this draft without
+    # opening the details eye. Only draft_prompt rides the heading; other
+    # experiments stay solely in the popover. draft_prompt ALSO keeps its full
+    # one-sentence meaning in the popover (unlike 'lane', which is skipped
+    # there) because the compact token alone doesn't say what the arm does.
+    arm = _compact_arm((d.get("experiments") or {}).get("draft_prompt"))
+    if arm:
+        bits.append(arm)
     return " · ".join(bits)
 
 
@@ -1571,7 +1618,19 @@ class _ReviewController(NSObject):
             pass
         if thread_url:
             thread_tv.setDelegate_(self)
-        _set_thread_text(thread_tv, thread_en or d.get("thread_text"), thread_url)
+        # Reddit (2026-07-17 user direction): the thread has a TITLE and a
+        # BODY (selftext). Render the title bold with the body beneath it in
+        # this same quote box. thread_text carries the title (see
+        # merge_review_queue._reddit_plan_to_candidates); thread_selftext is
+        # the body, empty for link posts / title-only threads (then this falls
+        # through to the plain title-only render). Twitter has no selftext, so
+        # it always takes the else branch and renders exactly as before.
+        _thread_title = thread_en or d.get("thread_text")
+        _thread_body = (d.get("thread_selftext") or "").strip() if _plat == "reddit" else ""
+        if _thread_body:
+            _set_thread_text(thread_tv, _thread_body, thread_url, title=_thread_title)
+        else:
+            _set_thread_text(thread_tv, _thread_title, thread_url)
         thread_scroll.setDocumentView_(thread_tv)
         content.addSubview_(thread_scroll)
         # Reply heading — bold. A concise "project/lane · viral N" tag rides

package/mcp/menubar/s4l_card_canvas.py

@@ -69,6 +69,7 @@ from AppKit import (
     NSWindowOcclusionStateVisible,
     NSBackingStoreBuffered,
     NSNormalWindowLevel,
+    NSFloatingWindowLevel,
     NSApplicationActivationPolicyAccessory,
     NSWindowStyleMaskTitled,
     NSWindowStyleMaskClosable,
@@ -247,13 +248,57 @@ class _CanvasController(NSObject):
         panel.orderFrontRegardless()
         self._log_surface("presented")
         if self._focus:
+            self.raise_to_front()
+
+    @objc.python_method
+    def raise_to_front(self):
+        """Reliably bring the canvas above the current front app ONCE. The panel
+        is a NON-activating panel at NSNormalWindowLevel (so it behaves like a
+        normal window the reviewer works IN, per 2026-07-16), which means a plain
+        makeKeyAndOrderFront on our accessory app can leave it BEHIND the
+        frontmost app: window-level ordering, not activation, decides what covers
+        what, and a normal-level window can't jump above another normal-level
+        window that belongs to the active app. That is exactly the 2026-07-17
+        Nhat report -- the canvas 'presented' occluded ("didn't surface"). Fix:
+        briefly promote to NSFloatingWindowLevel so orderFrontRegardless is
+        guaranteed to place it in front regardless of activation, then demote
+        back to normal a beat later so other apps can cover it again (keeping the
+        "work IN it like a normal window" posture; always-on-top is the corner
+        card's job, not this one)."""
+        panel = self._panel
+        if panel is None:
+            return
+        try:
+            NSApp.setActivationPolicy_(NSApplicationActivationPolicyAccessory)
+            NSApp.activateIgnoringOtherApps_(True)
+        except Exception:
+            pass
+        try:
+            panel.setLevel_(NSFloatingWindowLevel)
+        except Exception:
+            pass
+        try:
+            panel.makeKeyAndOrderFront_(None)
+            panel.orderFrontRegardless()
+        except Exception:
+            pass
+        # Drop back to a normal window level once it's on screen. performSelector
+        # keeps it above other windows just long enough to land in front; if
+        # scheduling ever fails, demote inline so it can never stay stuck on top.
+        try:
+            self.performSelector_withObject_afterDelay_("demoteWindowLevel:", None, 0.6)
+        except Exception:
             try:
-                NSApp.setActivationPolicy_(NSApplicationActivationPolicyAccessory)
-                NSApp.activateIgnoringOtherApps_(True)
+                panel.setLevel_(NSNormalWindowLevel)
             except Exception:
                 pass
-            panel.makeKeyAndOrderFront_(None)
-            panel.orderFrontRegardless()
+
+    def demoteWindowLevel_(self, _):
+        try:
+            if self._panel is not None:
+                self._panel.setLevel_(NSNormalWindowLevel)
+        except Exception:
+            pass
 
     @objc.python_method
     def _render(self):
@@ -457,6 +502,15 @@ class _CanvasController(NSObject):
         self._reflow_from(idx)
         self._resize_doc()
         self._refresh_header()
+        # Last card decided -> nothing left to review, so close the canvas
+        # instead of leaving an empty "0 pending drafts" window on screen.
+        # _finish() fires on_complete with every decision made this session,
+        # exactly like the corner card auto-finishing past its final draft
+        # (s4l_card._advance). Reached from a tile completing (approve/reject)
+        # and from prune_drafts draining the last candidate; either way an
+        # empty canvas has no reason to stay up.
+        if not self._order and self._panel is not None:
+            self._finish()
 
     # Approve All / Discard All buttons live in the header row (see
     # _render()), not a dedicated card -- these two actions just target the
@@ -746,13 +800,10 @@ def focus_active():
     if c is None or c._panel is None:
         return False
     try:
-        NSApp.setActivationPolicy_(NSApplicationActivationPolicyAccessory)
-        NSApp.activateIgnoringOtherApps_(True)
-    except Exception:
-        pass
-    try:
-        c._panel.makeKeyAndOrderFront_(None)
-        c._panel.orderFrontRegardless()
+        # Same reliable front-raise as the initial focused present: a brief
+        # window-level promotion so an already-open canvas comes above the front
+        # app even though it is a non-activating normal-level panel.
+        c.raise_to_front()
         c._panel.makeMainWindow()
         return True
     except Exception: