@m13v/s4l 1.6.197-rc.7

package/skill/engage-linkedin.sh

@@ -0,0 +1,581 @@
+#!/usr/bin/env bash
+# engage-linkedin.sh — LinkedIn engagement loop
+# Phase A: Discover replies/mentions from LinkedIn notifications (Claude-driven MCP).
+# Phase B: Respond to pending LinkedIn replies (Claude-driven, OAuth API for posting).
+# Called by launchd every 3 hours.
+#
+# IMPORTANT: all LinkedIn browser work goes through the linkedin-harness MCP
+# (bh_run, CDP-driven real Chrome on port 9556), driven by Claude (the LLM).
+# Do NOT re-introduce Python Playwright scrapers, Voyager API calls
+# (/voyager/api/*), comment-page scroll+expand loops, or programmatic re-login
+# flows. See CLAUDE.md "LinkedIn: flagged patterns to avoid" for why.
+
+set -euo pipefail
+
+# LinkedIn killswitch (2026-05-27): refuse to run if a prior fire detected
+# session compromise (http_999, authwall, throttle, li_at cleared).
+# State: ~/.claude/social-autoposter/linkedin.killswitch
+# Clear: python3 ~/social-autoposter/scripts/linkedin_killswitch.py clear
+if [ -f "$HOME/.claude/social-autoposter/linkedin.killswitch" ]; then
+    echo "[$(date +%H:%M:%S)] LINKEDIN_KILLSWITCH active. Aborting LinkedIn pipeline."
+    echo "  Re-auth LinkedIn in harness Chrome, then: python3 ~/social-autoposter/scripts/linkedin_killswitch.py clear"
+    exit 0
+fi
+
+# 2026-05-01: lock policy changed from "hold the entire run" to "hold only
+# while a Claude phase is actively driving the browser" — same pattern as
+# run-linkedin.sh. The old policy held linkedin-browser for the whole 25-45min
+# cycle, starving peer pipelines (run-linkedin, dm-replies-linkedin) and
+# defeating launchd schedules. The browser is only used inside the two
+# run_claude.sh invocations (Phase A discovery, Phase B reply). Everything
+# between (DB cleanup, pending pull, top performers, styles) is pure DB/CPU.
+source "$(dirname "$0")/lock.sh"
+# Browser backend bootstrap (linkedin-harness). Sets MCP_CONFIG_FILE,
+# BROWSER_INSTRUCTIONS, exports LINKEDIN_CDP_URL, and provides
+# ensure_linkedin_browser_for_backend. Migrated off the deprecated
+# linkedin-agent MCP on 2026-05-29 (mirrors the Twitter harness migration).
+source "$(dirname "$0")/lib/linkedin-backend.sh"
+
+# Load secrets
+# shellcheck source=/dev/null
+[ -f "$HOME/social-autoposter/.env" ] && source "$HOME/social-autoposter/.env"
+
+REPO_DIR="$HOME/social-autoposter"
+SKILL_FILE="$REPO_DIR/SKILL.md"
+LOG_DIR="$REPO_DIR/skill/logs"
+BATCH_SIZE=500
+MCP_CONFIG="$MCP_CONFIG_FILE"
+
+# DB-free since 2026-06-01: all reply state goes through the s4l.ai HTTP API
+# (X-Installation auth). No DATABASE_URL needed; the helpers below call the API.
+PY_BIN="$(command -v python3 || echo /usr/bin/python3)"
+
+# li_reply_count <status>  -> integer count of linkedin replies in that status.
+# Backed by GET /api/v1/replies/counts (same aggregate reply_db.py status uses).
+li_reply_count() {
+    "$PY_BIN" -c "
+import sys; sys.path.insert(0, '$REPO_DIR/scripts')
+from http_api import api_get
+resp = api_get('/api/v1/replies/counts', {'platform': 'linkedin'})
+counts = ((resp or {}).get('data') or {}).get('counts') or []
+want = '$1'
+print(next((int(r.get('count', 0)) for r in counts if r.get('status') == want), 0))
+" 2>/dev/null || echo 0
+}
+
+# li_reset_processing <older_than_hours>  -> count of rows reset to pending.
+# older_than_hours=0 means no time gate (reset every processing row). Backed by
+# POST /api/v1/replies/reset-stuck.
+li_reset_processing() {
+    "$PY_BIN" -c "
+import sys; sys.path.insert(0, '$REPO_DIR/scripts')
+from http_api import api_post
+resp = api_post('/api/v1/replies/reset-stuck', {'platform': 'linkedin', 'older_than_hours': int('$1')})
+print(((resp or {}).get('data') or {}).get('reset_count', 0))
+" 2>/dev/null || echo 0
+}
+
+mkdir -p "$LOG_DIR"
+LOG_FILE="$LOG_DIR/engage-linkedin-$(date +%Y-%m-%d_%H%M%S).log"
+
+# Per-cycle batch id stamped onto every claude_sessions row spawned by this
+# engagement run (via SA_CYCLE_ID env -> log_claude_session.py). 2026-05-10
+# cycle_id rollout.
+BATCH_ID="enli-$(date +%Y%m%d_%H%M%S)-$$"
+export SA_CYCLE_ID="$BATCH_ID"
+
+log() { echo "[$(date +%H:%M:%S)] $*" | tee -a "$LOG_FILE"; }
+
+RUN_START=$(date +%s)
+log "=== LinkedIn Engagement Run: $(date) ==="
+
+# Load exclusions from config
+EXCLUDED_AUTHORS=$(python3 -c "import json; c=json.load(open('$REPO_DIR/config.json')); print(', '.join(c.get('exclusions',{}).get('authors',[])))" 2>/dev/null || echo "")
+EXCLUDED_LINKEDIN=$(python3 -c "import json; c=json.load(open('$REPO_DIR/config.json')); print(', '.join(c.get('exclusions',{}).get('linkedin_profiles',[])))" 2>/dev/null || echo "")
+
+# ═══════════════════════════════════════════════════════
+# PHASE A: Discover new replies/mentions from LinkedIn notifications
+# Claude-driven: LLM navigates linkedin-agent MCP to /notifications/, extracts
+# actionable items from the notifications page DOM (NOT from Voyager API,
+# NOT by opening each permalink).
+# ═══════════════════════════════════════════════════════
+log "Phase A: Scanning LinkedIn notifications (Claude-driven)..."
+
+PHASE_A_PROMPT=$(mktemp)
+cat > "$PHASE_A_PROMPT" <<PROMPT_EOF
+You are the Social Autoposter LinkedIn discovery bot.
+
+Read $SKILL_FILE for content rules (tone, anti-AI detection, no em dashes).
+
+$BROWSER_INSTRUCTIONS
+
+## Task: Discover new LinkedIn replies and mentions from the notifications page
+
+CRITICAL - Browser agent rule: ONLY use the browser tool described in the BROWSER BACKEND block above (mcp__linkedin-harness__bh_run). NEVER use generic mcp__playwright-extension__*, mcp__isolated-browser__*, or mcp__macos-use__* tools.
+CRITICAL: If a browser tool call is blocked or times out, wait 30 seconds and retry. Repeat up to 3 times. If still blocked, stop.
+CRITICAL: Do NOT open individual post permalinks to fetch comment text. Everything we need is on the notifications page. Opening per-comment permalinks is a flagged scraping pattern.
+CRITICAL: Do NOT call any /voyager/api/ endpoint, do NOT fetch() from the linkedin.com session. Use only UI navigation (the navigate/snapshot/run-code equivalents in the BROWSER BACKEND block).
+
+EXCLUSIONS - do NOT engage with these accounts:
+- Excluded authors: $EXCLUDED_AUTHORS
+- Excluded LinkedIn profiles: $EXCLUDED_LINKEDIN
+- Skip comments by "Matthew Diakonov" or "m13v" (our own account).
+
+### Step 1: Load existing reply comment IDs (dedup)
+\`\`\`bash
+python3 ~/social-autoposter/scripts/li_discovery.py comment-ids
+\`\`\`
+Skip any notification whose comment URN is already in this list.
+
+### Step 2: Load author+post pairs we already engaged with
+\`\`\`bash
+python3 ~/social-autoposter/scripts/li_discovery.py engaged-pairs
+\`\`\`
+Skip any notification whose (author, post) pair is already here. One reply per author per thread.
+
+### Step 3: Load our LinkedIn posts for matching
+\`\`\`bash
+python3 ~/social-autoposter/scripts/li_discovery.py posts
+\`\`\`
+
+### Step 4: Navigate to LinkedIn notifications and verify session
+Navigate (per the BROWSER BACKEND block) to https://www.linkedin.com/notifications/
+
+Take a snapshot/screenshot and verify the page is the notifications feed (not a login/checkpoint page). If you see login, captcha, or a verification challenge, STOP immediately and print: SESSION_INVALID — do not attempt to log in. Exit.
+
+### Step 5: Load more notifications
+Scroll the page down a few times to lazy-load notifications. If a "Show more results" button is visible, click it — up to 5 times total, with a pause of 2-3 seconds between clicks. Stop if the button disappears.
+
+### Step 6: Extract actionable notifications from the notifications page DOM
+Run this DOM-extraction JS via the BROWSER BACKEND block's run-code equivalent
+(bh_run with js("""...""")). It is a single read of the already-loaded page — do
+NOT navigate to any other URL. The js() helper runs the body in page context and
+returns the result, so pass exactly this body (it ends with \`return\`):
+
+\`\`\`javascript
+const actionable = [];
+const actionablePhrases = [
+  'replied to your comment',
+  'mentioned you in a comment',
+  'mentioned you in this',
+  'commented on your post',
+  'commented on your update',
+];
+
+for (const article of document.querySelectorAll('article')) {
+  const text = (article.innerText || '').toLowerCase();
+  const matched = actionablePhrases.find(p => text.includes(p));
+  if (!matched) continue;
+
+  const strong = article.querySelector('strong');
+  const author = strong ? strong.textContent.trim() : 'unknown';
+
+  const link = article.querySelector('a[href*="commentUrn"]') ||
+               article.querySelector('a[href*="replyUrn"]') ||
+               article.querySelector('a[href*="feed/update"]');
+  const href = link ? link.getAttribute('href') : null;
+  if (!href) continue;
+
+  // Extract activity ID and commentUrn from the href
+  const activityMatch = href.match(/urn:li:activity:(\d+)/);
+  const activityId = activityMatch ? activityMatch[1] : null;
+  const commentUrnMatch = href.match(/commentUrn=([^&]+)/);
+  const commentUrn = commentUrnMatch ? decodeURIComponent(commentUrnMatch[1]) : null;
+
+  // Best-effort snippet: text inside the article, minus the author header
+  const snippet = (article.innerText || '').replace(/\s+/g, ' ').trim();
+
+  actionable.push({
+    type: matched,
+    author,
+    href: href.startsWith('http') ? href : ('https://www.linkedin.com' + href),
+    activity_id: activityId,
+    comment_urn: commentUrn,
+    snippet,
+  });
+}
+return JSON.stringify(actionable);
+\`\`\`
+
+### Step 7: For each extracted notification, decide whether to insert
+For each item:
+- If comment_urn is null OR activity_id is null: skip (no_comment_urn)
+- If comment_urn is in the Step 1 dedup list: skip (already_tracked)
+- If author matches an excluded account or is our own: skip (excluded_author / own_account)
+- Build author_post_key = author + '|||' + our_url-for-this-post. If this pair is in the Step 2 list: skip (author_already_engaged)
+- Find matching post_id from Step 3 by activity_id in the our_url. If none: create one (use PROJECT_NAME matched from config.json projects[].topics against the post topic):
+\`\`\`bash
+python3 ~/social-autoposter/scripts/li_discovery.py create-post --activity-id "ACTIVITY_ID" --project "PROJECT_NAME" --author "AUTHOR"
+\`\`\`
+This prints the post id (reusing the existing row on duplicate). Use that id as POST_ID below.
+
+Insert the reply:
+\`\`\`bash
+python3 ~/social-autoposter/scripts/li_discovery.py insert-reply --post-id POST_ID --comment-urn "COMMENT_URN" --author "AUTHOR" --content "SNIPPET" --href "HREF"
+\`\`\`
+Prints the new reply id, or "duplicate" / "gated:<reason>" (both are non-fatal, just move to the next item).
+
+### Step 8: Summary
+Print:
+- N new replies discovered
+- N already tracked
+- N author already engaged on thread
+- N excluded
+- N own account
+- N no comment URN
+
+### Step 8b: Structured scan-summary marker line (REQUIRED)
+After the human-readable summary above, print EXACTLY ONE line with this
+format (no other text on the line, all values are integers):
+
+LINKEDIN_SCAN_SUMMARY: scanned=<TOTAL_NOTIFICATIONS_INSPECTED> new=<NEW_INSERTED> already=<ALREADY_TRACKED> excluded=<EXCLUDED+OWN_ACCOUNT> unmatched=<NO_COMMENT_URN>
+
+The wrapper shell script greps this line to surface scan-stage pills
+(scanned / new / excluded) on the dashboard Result column. If a counter
+doesn't apply this run, emit it as 0 anyway. The line MUST start with
+"LINKEDIN_SCAN_SUMMARY:" at column 0 with that exact spelling and casing.
+PROMPT_EOF
+
+# Acquire linkedin-browser ONLY around the Phase A Claude run. lock.sh is
+# FIFO-queued, so a peer pipeline (run-linkedin, dm-replies-linkedin) that's
+# mid-run blocks here rather than skipping. run_claude.sh auto-exports
+# SA_PIPELINE_LOCKED=1 + SA_PIPELINE_PLATFORM so the PreToolUse hook
+# (~/.claude/hooks/linkedin-agent-lock.sh) skips the cross-session block check.
+acquire_lock "linkedin-browser" 3600
+ensure_linkedin_browser_for_backend 2>&1 | tee -a "$LOG_FILE"
+
+gtimeout 1800 "$REPO_DIR/scripts/run_claude.sh" "engage-linkedin-phaseA" --strict-mcp-config --mcp-config "$MCP_CONFIG" --output-format stream-json --verbose -p "$(cat "$PHASE_A_PROMPT")" 2>&1 | tee -a "$LOG_FILE" || log "WARNING: Phase A claude exited with code $?"
+
+release_lock "linkedin-browser"
+# Defense-in-depth: explicitly clear the hook-layer lockfile so the next
+# pipeline cycle's PreToolUse never sees a stale entry from us. The
+# run_claude.sh exit trap already does this in the happy path; this
+# repeat is harmless and covers SIGKILL of run_claude.sh.
+rm -f "$HOME/.claude/linkedin-agent-lock.json"
+rm -f "$PHASE_A_PROMPT"
+
+# ═══════════════════════════════════════════════════════
+# PHASE B: Respond to pending LinkedIn replies
+# Claude-driven. Posts via OAuth API (api.linkedin.com/v2/socialActions) by
+# default (documented, authorized integration). Falls back to linkedin-agent
+# MCP browser click-through if API errors.
+# ═══════════════════════════════════════════════════════
+
+# Reset any 'processing' items older than 2 hours back to 'pending'
+RESET_COUNT=$(li_reset_processing 2)
+[ "$RESET_COUNT" -gt 0 ] && log "Phase B: Reset $RESET_COUNT stuck 'processing' LinkedIn items back to pending"
+
+PENDING_COUNT=$(li_reply_count pending)
+
+if [ "$PENDING_COUNT" -eq 0 ]; then
+    log "Phase B: No pending LinkedIn replies. Done!"
+else
+    log "Phase B: $PENDING_COUNT pending LinkedIn replies to process"
+
+    # Pull the full pending batch via the next-pending endpoint (no DATABASE_URL).
+    # The route applies the same ordering the old json_agg query used (our own
+    # posts first, then oldest discovered_at) and returns {replies:[...]}. We
+    # extract the array so PENDING_DATA stays the same JSON shape the prompt
+    # consumed before the migration.
+    PENDING_DATA=$("$PY_BIN" -c "
+import json, sys
+sys.path.insert(0, '$REPO_DIR/scripts')
+from http_api import api_get
+resp = api_get('/api/v1/replies/next-pending', {'platform': 'linkedin', 'limit': int('$BATCH_SIZE')})
+rows = (resp.get('data') or {}).get('replies') or []
+print(json.dumps(rows))
+" 2>/dev/null || echo "[]")
+
+    # Per-project voice map (so each reply can be drafted in the matched project's voice)
+    PROJECTS_VOICE_JSON=$(python3 -c "
+import json
+c = json.load(open('$REPO_DIR/config.json'))
+print(json.dumps({p['name']: p.get('voice', {}) for p in c.get('projects', []) if p.get('voice')}, indent=2))
+" 2>/dev/null || echo "{}")
+
+    # Engagement-style picker (2026-05-31 LinkedIn alignment to Twitter): pick
+    # ONE assigned style per reply iteration PROGRAMMATICALLY, mirroring
+    # engage-twitter.sh. The picked style flows two places: (1) --style filter
+    # for top_performers.py so the per-style exemplars match the assignment,
+    # (2) saps_render_style_block so the prompt embeds the same assignment. On
+    # invent mode picked_style is empty and top_performers stays unfiltered.
+    # Replaces the legacy generate_styles_block (which discarded the pick and
+    # let the model invent freely).
+    source "$REPO_DIR/skill/styles.sh"
+    STYLE_ASSIGN_FILE=$(mktemp -t saps_linkedin_eng_assign_XXXXXX.json)
+    saps_pick_style linkedin replying "$STYLE_ASSIGN_FILE" >/dev/null 2>&1 || true
+    PICKED_STYLE=$(python3 -c "
+import json
+try:
+    with open('$STYLE_ASSIGN_FILE') as f:
+        d = json.load(f)
+    print(d.get('style') or '')
+except Exception:
+    print('')
+" 2>/dev/null)
+    PICKED_MODE=$(python3 -c "
+import json
+try:
+    with open('$STYLE_ASSIGN_FILE') as f:
+        d = json.load(f)
+    print(d.get('mode') or 'use')
+except Exception:
+    print('use')
+" 2>/dev/null)
+    STYLES_BLOCK=$(saps_render_style_block "$STYLE_ASSIGN_FILE" linkedin replying)
+    rm -f "$STYLE_ASSIGN_FILE" 2>/dev/null || true
+
+    # Top performers feedback report — filtered to the picked style when in
+    # 'use' mode so the few-shot exemplars match the assignment.
+    if [ -n "$PICKED_STYLE" ]; then
+        TOP_REPORT=$(python3 "$REPO_DIR/scripts/top_performers.py" --platform linkedin --style "$PICKED_STYLE" 2>/dev/null || echo "(top performers report unavailable)")
+    else
+        TOP_REPORT=$(python3 "$REPO_DIR/scripts/top_performers.py" --platform linkedin 2>/dev/null || echo "(top performers report unavailable)")
+    fi
+
+    PHASE_B_PROMPT=$(mktemp)
+    cat > "$PHASE_B_PROMPT" <<PROMPT_EOF
+You are the Social Autoposter LinkedIn engagement bot.
+
+Read $SKILL_FILE for the full workflow, content rules, and platform details.
+
+$BROWSER_INSTRUCTIONS
+
+EXCLUSIONS - do NOT engage with these accounts (skip and mark as 'skipped' with reason 'excluded_author'):
+- Excluded authors: $EXCLUDED_AUTHORS
+- Excluded LinkedIn profiles: $EXCLUDED_LINKEDIN
+
+### BOT / ENGAGEMENT-LOOP ESCAPE HATCH (use sparingly, but use it)
+We maintain a universal author blocklist in Postgres (\`author_blocklist\`),
+consulted at /api/v1/replies POST time. A single block recorded by ANY of
+our accounts/installs applies to EVERY future engagement from EVERY of our
+accounts — universal scope, by design. The velocity gate already covers
+"this handle has gotten too many replies from us in 24h/7d"; this lane is
+for the LLM-judgment cases velocity cannot catch.
+
+When to add a block (your judgment, exercised CONSERVATIVELY):
+- The profile is plainly an AI/bot account: templated phrasing, generic
+  filler answers, name/headline reads "AI growth agent" / "comments on
+  posts for you", bio is engagement-farming boilerplate
+- We are clearly stuck in a reciprocal engagement loop with this profile
+  (they comment on every one of our posts, we reply to every one of theirs,
+  no substance is exchanged)
+- The profile is engagement farming (mass low-effort comments across the
+  platform, not actually engaging with the topic)
+
+DO NOT add a block for: someone we disagree with, a hostile-but-human
+critic, a low-quality but human comment, or a single bad interaction.
+Skip those (status='skipped') — blocking is permanent until manually
+removed and applies to all our accounts.
+
+How to add the block (run BEFORE marking the current reply skipped). The
+handle to pass is the URL-vanity portion of /in/<vanity>/ (e.g. for
+linkedin.com/in/jane-doe-123/, pass jane-doe-123):
+  python3 \$REPO_DIR/scripts/reply_db.py blocklist add linkedin HANDLE \\
+    --reason "<one-line judgment>" \\
+    --classification {bot|engagement_loop} \\
+    --source-reply-id REPLY_ID
+
+Then mark the current reply skipped with a clear reason:
+  python3 \$REPO_DIR/scripts/reply_db.py skipped REPLY_ID "blocklist_added:HANDLE"
+
+CRITICAL - Browser agent rule: ONLY use the browser tool described in the BROWSER BACKEND block above (mcp__linkedin-harness__bh_run). NEVER use generic mcp__playwright-extension__*, mcp__isolated-browser__*, or mcp__macos-use__* tools.
+CRITICAL: If a browser tool call is blocked or times out, DO NOT fall back to any other browser tool. Wait 30 seconds and retry. Repeat up to 3 times.
+CRITICAL: TECHNICAL FAILURES ARE NOT TERMINAL. If after retries the action still failed for any technical reason (browser blocked, MCP timeout, page rendering issue, linkedin.com unreachable, linkedin_api.py 5xx), DO NOT call reply_db.py skipped. Leave the row in 'processing' status and move on to the next pending item. The next engage run's start-of-script cleanup resets stuck 'processing' rows back to 'pending' and retries automatically.
+CRITICAL: ONLY call reply_db.py skipped for content/policy reasons (e.g., light_acknowledgment, drive_by_self_promo, hostile_user, off_topic, troll, excluded_author). NEVER skip for technical browser/network failures: those must be retry-able.
+CRITICAL: Do NOT call /voyager/api/ endpoints. Posting goes through linkedin_api.py (OAuth api.linkedin.com). Browser is the fallback only.
+
+## Respond to pending LinkedIn replies ($PENDING_COUNT total)
+
+### Priority order:
+1. **Replies on our original posts** (is_our_original_post=1) - highest priority
+2. **Direct questions** ("what tool", "how do you", "can you share")
+3. **Everything else** - general engagement
+
+### Tiered link strategy:
+- **Tier 1 (default):** No link. Genuine engagement, expand topic.
+- **Tier 2 (natural mention):** Conversation touches a topic matching a project in config. Recommend it casually as a tool you've come across.
+- **Tier 3 (direct ask):** They ask for link/tool/source. Give it immediately.
+
+## FEEDBACK FROM PAST PERFORMANCE (use this to write better replies):
+$TOP_REPORT
+
+$STYLES_BLOCK
+
+## Per-project voice map
+For each reply you draft, look up the matched project's voice block below and apply it: follow \`voice.tone\`, never violate any item in \`voice.never\`, mirror \`voice.examples\` / \`voice.examples_good\` when present.
+$PROJECTS_VOICE_JSON
+
+## Resolving the parent post (replaces the old prompt-blob index)
+Each pending row's \`project_name\` is a best-effort guess. After navigating the thread (Step 2), extract the activity_id from the page URL/comment URN and resolve it via:
+  python3 $REPO_DIR/scripts/lookup_post.py linkedin ACTIVITY_ID
+Returns JSON: {"project": "fazm", "our_content": "...full text...", "thread_url": "..."} or {"project": null} if it's not one of our posts.
+
+Here are the replies to process:
+$PENDING_DATA
+
+CRITICAL: Reply in the SAME LANGUAGE as the message you are responding to. Match the language exactly.
+CRITICAL: Process EVERY reply. For each: either post a response and mark as 'replied', OR mark as 'skipped' with a skip_reason.
+
+CRITICAL: For ALL database operations, use the reply_db.py helper (NOT raw psql):
+  python3 $REPO_DIR/scripts/reply_db.py processing ID          # BEFORE posting
+  python3 $REPO_DIR/scripts/reply_db.py replied ID "reply text" [url] [engagement_style] [is_recommendation]   # AFTER posting. engagement_style is TONE (critic, storyteller, etc). Pass "1" for is_recommendation ONLY when the reply casually recommends a project (Tier 2/3); leave blank otherwise.
+  python3 $REPO_DIR/scripts/reply_db.py skipped ID "reason"
+  python3 $REPO_DIR/scripts/reply_db.py skip_batch '{"ids":[1,2,3],"reason":"..."}'
+  python3 $REPO_DIR/scripts/reply_db.py status
+NEVER use psql directly for reply status updates.
+
+### Project tracking on replies
+When you recommend a project in a reply (Tier 2 or Tier 3), set project_name on the reply:
+  python3 $REPO_DIR/scripts/reply_db.py set_project REPLY_ID "PROJECT_NAME"
+
+MANDATORY reply flow for every item:
+  Step 1: python3 reply_db.py processing ID      <- mark BEFORE posting
+  Step 2: NAVIGATE TO THE THREAD AND READ CONTEXT (mandatory, do NOT skip).
+          Do NOT draft a reply from the notification snippet alone — the snippet
+          is truncated and lacks the parent post content + sibling replies.
+          a) Navigate (per the BROWSER BACKEND block) to their_comment_url
+          b) Snapshot/screenshot (per the BROWSER BACKEND block) to read:
+             - the FULL parent post text (our original post if this is on our thread)
+             - the immediate ancestor of their_comment_id
+             - sibling replies (so you don't repeat what someone else already said)
+          c) Extract the activity_id (the long numeric string after \`urn:li:activity:\`)
+             from the URL or comment URN. Resolve it:
+               python3 $REPO_DIR/scripts/lookup_post.py linkedin ACTIVITY_ID
+             If the response has a non-null \"project\", that's our post, OVERRIDE
+             the reply row and use that project's voice for drafting:
+               python3 $REPO_DIR/scripts/reply_db.py set_project REPLY_ID "RESOLVED_PROJECT"
+             Use the returned \"our_content\" as the FULL text of the post being
+             replied to (more accurate than the truncated our_content in PENDING_DATA).
+             If \"project\" is null, we're a guest in someone else's thread; keep
+             the existing project_name and follow global content rules.
+  Step 3: Draft the reply using the resolved project's voice + the ASSIGNED
+          engagement style. This cycle: mode=$PICKED_MODE style='${PICKED_STYLE:-(invent)}'.
+          In USE mode ($PICKED_MODE=use) apply the assigned style '${PICKED_STYLE}'
+          verbatim; do NOT pick a different style and do NOT invent one. In INVENT
+          mode ($PICKED_MODE=invent) craft a NEW snake_case style name not in the
+          curated block above and pass it as the [engagement_style] arg in Step 5.
+          Professional but casual. NEVER em dashes. Match parent post language.
+  Step 4: post reply (OAuth API first, browser fallback)
+  Step 5: python3 reply_db.py replied ID "text" [url] [engagement_style] [is_recommendation]   <- mark AFTER success. engagement_style is the style name you applied (in USE mode the assigned '${PICKED_STYLE}'). is_recommendation="1" only when you mentioned a project (Tier 2/3).
+If Step 5 fails, the item stays 'processing' and will be reset to 'pending' on the next run.
+
+For LinkedIn replies - use the OAuth API first:
+1. Extract the activity ID from their_comment_url or their_comment_id.
+   - From their_comment_id like \`urn:li:comment:(activity:7438226125077549056,7438815640536170496)\`, the activity ID is \`7438226125077549056\` and the full URN is the parent_comment_urn.
+   - From their_comment_url, extract the activity ID from the URL path.
+2. Post the reply via API:
+   \`\`\`bash
+   python3 $REPO_DIR/scripts/linkedin_api.py reply ACTIVITY_ID "PARENT_COMMENT_URN" "YOUR REPLY TEXT" --project "RESOLVED_PROJECT_NAME" --reply-id REPLY_ID
+   \`\`\`
+   Replace RESOLVED_PROJECT_NAME with the project from lookup_post.py (or the row's project_name if lookup returned null).
+   Replace REPLY_ID with the numeric id from the current reply row.
+   The API will automatically wrap any URLs in the reply text with short tracking links and backfill the reply_id after posting.
+   This returns JSON with {ok, reply_urn, permalink}. Use permalink as the reply URL.
+   On {"ok": true}: skip step 3 (browser fallback) and skip the browser-based
+   verification in step 5 below — the API success response (with reply_urn) is
+   itself authoritative. Mark replied with the permalink. Do NOT navigate the
+   browser to verify; that would burn the linkedin-browser lock for no gain.
+3. If the API call fails (e.g., token expired, comment deleted), fall back to the linkedin-agent browser:
+   - UTM-WRAP YOUR_REPLY_TEXT FIRST. The browser-typing path has no Python wrap
+     layer, so a bare URL would be posted as-is and we'd lose attribution. Run:
+       python3 \$REPO_DIR/scripts/dm_short_links.py utm-text \\
+         --platform linkedin \\
+         --project RESOLVED_PROJECT_NAME \\
+         --text "YOUR_REPLY_TEXT"
+     Use the printed string going forward. No DB write happens. Safe to run
+     unconditionally (no-op when YOUR_REPLY_TEXT contains zero URLs).
+   - Navigate to their_comment_url (per the BROWSER BACKEND block)
+   - Snapshot/screenshot to find the comment, click Reply (click_at_xy on the Reply control), type the (UTM-wrapped) text, submit
+   - Do NOT aggressively scroll-and-expand comments; if the comment isn't visible after a normal scroll, mark as 'skipped' with reason 'comment_not_found'
+4. If both API and browser fail, mark as 'skipped' with reason 'comment_not_found'.
+
+5. POST-SUBMIT VERIFICATION (mandatory, BROWSER-FALLBACK PATH ONLY).
+   If you posted via the OAuth API in step 2 and got {"ok": true}, SKIP this
+   block entirely — the API response is the verification. Run this block ONLY
+   when step 3's browser fallback was used. Verify visually + via DOM (the
+   harness has no network-capture tool; do NOT try to read /voyager or socialActions
+   network traffic — that is a flagged pattern anyway):
+   5a. Capture a screenshot (bh_run print(capture_screenshot())) and Read the PNG to
+       check for a toast / submit feedback.
+   5b. Read the DOM via the BROWSER BACKEND block's run-code equivalent (bh_run with
+       js("""...""")) and check:
+         (a) a fresh comment by 'Matthew Diakonov' / 'You' is rendered under their_comment
+         (b) NO 'could not be created' / 'try again' / 'something went wrong' toast text
+         (c) reply editor textbox cleared (empty contenteditable)
+   5c. SUCCESS = (a) passes with no toast. REJECTED = error toast present OR our
+       reply not visible in the DOM.
+6. If REJECTED, do NOT call reply_db.py replied. Mark soft-blocked:
+     python3 $REPO_DIR/scripts/reply_db.py skipped ID "soft_blocked: <verbatim toast or 'quiet_fail_count_unchanged'>"
+   Then STOP this row and move to the next pending reply.
+7. If step 5 SUCCESS (or step 2 OAuth success), mark replied via Step 5 of the
+   MANDATORY reply flow above (reply_db.py replied ID "text" [url] [style] [is_recommendation]).
+
+After every 10 replies, run: python3 $REPO_DIR/scripts/reply_db.py status
+PROMPT_EOF
+
+    # Re-acquire linkedin-browser ONLY for Phase B. The lock was released after
+    # Phase A so peer pipelines could use the browser during our DB-pull /
+    # styles-prep window (~1-3s). FIFO ticket queue in lock.sh ensures
+    # fairness if a peer or parallel cycle grabbed it in the meantime.
+    acquire_lock "linkedin-browser" 3600
+    ensure_linkedin_browser_for_backend 2>&1 | tee -a "$LOG_FILE"
+
+    gtimeout 5400 "$REPO_DIR/scripts/run_claude.sh" "engage-linkedin-phaseB" --strict-mcp-config --mcp-config "$MCP_CONFIG" --output-format stream-json --verbose -p "$(cat "$PHASE_B_PROMPT")" 2>&1 | tee -a "$LOG_FILE" || log "WARNING: Phase B claude exited with code $?"
+
+    release_lock "linkedin-browser"
+    # Defense-in-depth: explicit hook-lockfile cleanup; see Phase A note.
+    rm -f "$HOME/.claude/linkedin-agent-lock.json"
+    rm -f "$PHASE_B_PROMPT"
+fi
+
+# Reset any items left in 'processing' after subprocess exit (tech-failure
+# retry path: agent leaves rows here on browser/MCP failure rather than
+# calling reply_db.py skipped, so the next run picks them up automatically).
+# older_than_hours=0 resets ALL 'processing' rows regardless of age.
+POST_RESET=$(li_reset_processing 0)
+[ "$POST_RESET" -gt 0 ] && log "Post-run: Reset $POST_RESET 'processing' LinkedIn items back to pending"
+
+# ═══════════════════════════════════════════════════════
+# Cleanup
+# ═══════════════════════════════════════════════════════
+TOTAL_PENDING=$(li_reply_count pending)
+TOTAL_REPLIED=$(li_reply_count replied)
+TOTAL_SKIPPED=$(li_reply_count skipped)
+
+log "LinkedIn summary: pending=$TOTAL_PENDING replied=$TOTAL_REPLIED skipped=$TOTAL_SKIPPED"
+
+# Log run to persistent monitor
+RUN_ELAPSED=$(( $(date +%s) - RUN_START ))
+_COST=$(python3 "$REPO_DIR/scripts/get_run_cost.py" --since "$RUN_START" --scripts "engage-linkedin-phaseA" "engage-linkedin-phaseB" 2>/dev/null || echo "0.0000")
+# Pull Phase A scan-stage counters from the structured marker line Claude emits
+# at the end of Phase A: "LINKEDIN_SCAN_SUMMARY: scanned=N new=N already=N
+# excluded=N unmatched=N". If the marker is missing (Claude failed before
+# printing or the prompt drifted), fall back to no scan= segment and the
+# dashboard renders the old way.
+LI_SCAN_LINE=$(grep -m1 "^LINKEDIN_SCAN_SUMMARY:" "$LOG_FILE" 2>/dev/null || true)
+LI_SCAN_ARG=""
+if [ -n "$LI_SCAN_LINE" ]; then
+  li_scanned=$(echo "$LI_SCAN_LINE" | grep -oE "scanned=[0-9]+" | head -1 | cut -d= -f2)
+  li_new=$(echo "$LI_SCAN_LINE" | grep -oE "new=[0-9]+" | head -1 | cut -d= -f2)
+  li_already=$(echo "$LI_SCAN_LINE" | grep -oE "already=[0-9]+" | head -1 | cut -d= -f2)
+  li_excl=$(echo "$LI_SCAN_LINE" | grep -oE "excluded=[0-9]+" | head -1 | cut -d= -f2)
+  li_unm=$(echo "$LI_SCAN_LINE" | grep -oE "unmatched=[0-9]+" | head -1 | cut -d= -f2)
+  parts=""
+  [ -n "$li_scanned" ] && parts="${parts}scanned=${li_scanned},"
+  [ -n "$li_new" ]     && parts="${parts}new=${li_new},"
+  [ -n "$li_already" ] && [ "$li_already" -gt 0 ] && parts="${parts}already=${li_already},"
+  [ -n "$li_excl" ]    && [ "$li_excl"    -gt 0 ] && parts="${parts}excluded=${li_excl},"
+  [ -n "$li_unm" ]     && [ "$li_unm"     -gt 0 ] && parts="${parts}unmatched=${li_unm},"
+  LI_SCAN_ARG="${parts%,}"
+fi
+if [ -n "$LI_SCAN_ARG" ]; then
+  python3 "$REPO_DIR/scripts/log_run.py" --script "engage_linkedin" --posted "$TOTAL_REPLIED" --skipped "$TOTAL_SKIPPED" --failed 0 --cost "$_COST" --elapsed "$RUN_ELAPSED" --scan "$LI_SCAN_ARG"
+else
+  python3 "$REPO_DIR/scripts/log_run.py" --script "engage_linkedin" --posted "$TOTAL_REPLIED" --skipped "$TOTAL_SKIPPED" --failed 0 --cost "$_COST" --elapsed "$RUN_ELAPSED"
+fi
+
+# Delete old logs
+find "$LOG_DIR" -name "engage-linkedin-*.log" -mtime +7 -delete 2>/dev/null || true
+
+log "=== LinkedIn engagement complete: $(date) ==="

package/skill/engage-moltbook.sh

@@ -0,0 +1,36 @@
+#!/usr/bin/env bash
+# engage-moltbook.sh — MoltBook reply engagement loop
+# Calls engage_reddit.py --platform moltbook to process pending MoltBook replies.
+# Discovery runs separately via run-scan-moltbook-replies.sh.
+# Called by launchd every 10 minutes.
+
+set -euo pipefail
+
+source "$(dirname "$0")/lock.sh"
+acquire_lock "engage-moltbook" 3600
+
+[ -f "$HOME/social-autoposter/.env" ] && source "$HOME/social-autoposter/.env"
+
+# Cycle ID for cross-cycle cost accounting (see engage.sh / run-reddit-search.sh
+# for the same pattern). engage_reddit.py's claude subprocess inherits this via
+# env, and log_claude_session.py stamps claude_sessions.cycle_id.
+BATCH_ID="${BATCH_ID:-enmb-$(date +%Y%m%d-%H%M%S)}"
+export BATCH_ID
+export SA_CYCLE_ID="$BATCH_ID"
+
+REPO_DIR="$HOME/social-autoposter"
+LOG_DIR="$REPO_DIR/skill/logs"
+mkdir -p "$LOG_DIR"
+LOG_FILE="$LOG_DIR/engage-moltbook-$(date +%Y-%m-%d_%H%M%S).log"
+
+log() { echo "[$(date +%H:%M:%S)] $*" | tee -a "$LOG_FILE"; }
+
+RUN_START=$(date +%s)
+log "=== MoltBook Engage Run: $(date) ==="
+
+python3 "$REPO_DIR/scripts/engage_reddit.py" --platform moltbook 2>&1 | tee -a "$LOG_FILE" || log "WARNING: engage_reddit.py exited non-zero"
+
+RUN_ELAPSED=$(( $(date +%s) - RUN_START ))
+log "=== MoltBook Engage complete: $(date) (elapsed ${RUN_ELAPSED}s) ==="
+
+find "$LOG_DIR" -name "engage-moltbook-*.log" -mtime +7 -delete 2>/dev/null || true