Agent Fleet Architecture

Agents grouped by frequency

1. Hourly
   1. Digest QA (digest-qa): Autonomous QA of fleet digest emails, verifies claims against actual data, auto-fixes small issues, emails summary with audit trail Runs hourly.
   2. Error Bot (error-bot): Hourly production reliability, detect, triage, auto-fix errors Runs hourly.
   3. Follow-up Runner (follow-up-runner): Hourly executor for scheduled follow-up tasks from interactive sessions. Scans tools/follow-ups/pending/ and runs any task whose run_after has passed. Runs hourly.
   4. Skill Reflector (skill-reflector): Self-recursive skill/rule improvement, collects correction signals from transcripts, proposes edits, handles approval loop via email Runs hourly.
2. Daily, multiple slots
   1. CFO Spike Check (cfo-spike): Midweek cost spike detection, pure Python, no LLM Runs multiple times daily.
   2. Fleet Doctor (fleet-doctor): Autonomous anomaly investigation, classification, and remediation Runs multiple times daily.
   3. Flo AI Response Audit (flo-audit): Monitors Flo support chatbot responses for factual errors, wrong prices, fake features, and escalation events Runs multiple times daily.
   4. Indexing Monitor (indexing-monitor): Submit new URLs to Google IndexNow + inspect indexing status Runs multiple times daily.
   5. Session Digest (session-digest): Intraday PostHog session replay analysis (3x daily) Runs multiple times daily.
3. Daily
   1. A/B Test Monitor (ab-test-monitor): Daily statistical analysis of active experiments with chi-squared, sequential testing, and SRM detection. Runs daily 09:30.
   2. Analytics Digest (analytics-digest): Daily deep-dive into PostHog + GSC metrics with investigation Runs daily 01:15.
   3. Blacklist Monitor (blacklist-monitor): Daily DNS check of active sending domains against major blocklists (Spamhaus DBL/ZEN, SURBL, Barracuda, SpamCop, SORBS). Silent on clean runs; raises critical fleet alert on any hit. Domain list auto-synced from Smartlead. Runs daily 06:15.
   4. Commit Digest (commit-digest): Daily git log summarization, what shipped yesterday Runs daily 00:30.
   5. Crawl Health Monitor (crawl-health): Daily crawl-to-traffic funnel analysis, Vercel logs + inspection + GSC merged per segment Runs daily 02:00.
   6. CVG Portugal pilot digest (pilot-pt-digest): Daily roster + feedback summary during the CVG pilot window (2026-05-04 to 2026-06-08). Self-gates outside the window. Runs daily 07:00.
   7. Daily Cost Collection (cost-collection): Collects cost data from 17 providers into daily JSON snapshots Runs daily 06:00.
   8. DMARC Monitor (dmarc-monitor): Daily DMARC aggregate report ingestion and analysis. Fetches reports from flip inbox, parses XML, checks DKIM/SPF alignment per domain (DFY outbound + flipeducation.ai). Raises fleet alerts on authentication failures or alignment degradation. Runs daily 06:30.
   9. Evening Recap (evening-recap): Retrospective daily brief at 23:30. What happened today. Pairs with Morning Brief. Phase 4.3. Runs daily 23:30.
   10. Fleet Health Monitor (fleet-health): Daily fleet-wide health check, SLA, credentials, disk, LaunchAgent status Runs daily 08:00.
   11. GSC Archive (gsc-archive): Daily Google Search Console data archival for historical analysis Runs daily 07:00.
   12. Metric Anomaly Detector (metric-anomalies): Daily L1 sweep that catches anomalies on important business metrics: mission-q1 negative rate (with EN/non-EN split), generation success rate, signup conversion, JS exception rate. Compares last-7d to a 14-90d baseline (excludes recent 7-14d so multi-week shocks don't dilute themselves). Silent on green. Phase 4.2 G6, built after the W17 mission-feedback spike went unnoticed. Runs daily 06:30.
   13. Morning Brief Compiler (morning-brief): Daily consolidated email, analytics + errors + commits from overnight agents Runs daily 07:45.
   14. Morning Brief v2 (morning-brief-v2): Forward-looking daily brief at 07:00. Overnight escalations, decision queue, fleet snapshot. Phase 4.3. Runs in parallel with morning-brief during 1-week canary. Runs daily 07:00.
   15. PDF Email-Gate Heartbeat (pdf-gate-heartbeat): Daily synthetic deliverability probe for the PDF email-gate, POSTs a fixed test address + staging mission slug through deliverPdfEmail() and CRIT-alerts if Resend refuses the send Runs daily 06:00.
   16. Security Sentinel (security-sentinel): Machine-wide security monitoring, npm supply chain, secrets scanning, credential rotation, LaunchAgent integrity, macOS posture, web threat intelligence Runs daily 06:30.
   17. SEO Regression Monitor (seo-regression): Detects GSC ranking drops using adaptive IQR thresholds. No LLM. Runs daily 08:00.
   18. User Lifecycle Agent (user-lifecycle): Daily funnel tracking, visit → signup → mission → return → checkout Runs daily 07:30.
   19. Wiki Librarian (wiki-librarian): Compiles agent deliverables into atomic wiki entries; maintains INDEX.md and per-agent wiki digests Runs daily 01:00.
4. Interval
   1. Campaign Launch Monitor (campaign-monitor): Monitors active Smartlead campaigns for deliverability issues. Auto-pauses at 3% bounce. Checks every 10 minutes. Graduates campaigns after 24h clean sending. Runs every interval.
   2. Catch-Up Orchestrator (catch-up): Detects and re-runs missed scheduled tasks Runs every interval.
   3. Industry Scout (industry-scout): Monitors edtech industry news, competitor moves, market trends Runs every interval.
   4. Smartlead Token Refresh (smartlead-token-refresh): Rotates Gmail OAuth access tokens for 12 Smartlead mailboxes via DWD impersonation (45-min interval) Runs every interval.
   5. Tech Scout (tech-scout): Daily tech intelligence, stack impact + business-fit framing Runs every interval.
5. Weekly, selected days
   1. Data Analyst (data-analyst): Self-directed data quality, anomaly detection, and insight extraction across PostHog + Stripe + Supabase. Maintains investigation journal across runs. Runs selected weekdays 04:00.
6. Weekly
   1. Adriana Weekly (adriana-weekly): Monday consolidated email for Adriana, teacher stories, content needs, product insights Runs monday 09:00.
   2. Backup Integrity (backup-integrity): Verifies R2, Supabase, git, SQLite, LaunchAgent health. No LLM. Runs sunday 06:00.
   3. Beta Intelligence (beta-intelligence): Weekly user intelligence from micro-surveys, support chat, and behavioral telemetry, teacher profiles, feedback themes, conversion signals Runs Fri 10:00.
   4. CFO Agent (cfo): Weekly cost digest + midweek spike check across all providers Runs monday 08:00.
   5. Codex Fleet Canary (codex-fleet-canary): Read-only subscription-backed Codex canary for fleet-ops model evaluation Runs thursday 23:30.
   6. Competitive Landscape Synthesis (competitive-landscape): Weekly L3. Reads tech-scout + industry-scout digests over 7 days, dedups by URL, ranks by impact, emits unified competitive view. No LLM, no new collection, no scoring (existing scout scoring trusted). Phase 4.2 G4. Runs monday 11:00.
   7. Content Freshness Agent (content-freshness): Weekly scan for stale blog content, declining GSC performance, content gaps Runs tuesday 09:00.
   8. Dependency Security Audit (dep-audit): Weekly npm vulnerability scan. Auto-creates PR for safe patch bumps. Runs saturday 04:00.
   9. Emergent Signals Detector (emergent-signals): Weekly L3 digest, finds the unknown unknowns: new referrer domains, locale × methodology breakouts >2σ from baseline, new GSC queries climbing, country-level conversion surprises. Closes the Blackboard-traffic-class blind spot. Phase 4.2 G5. Runs sunday 22:00.
   10. Founder Cost Collection (founder-costs): Weekly Gmail invoice scan + manual entry merge for personal-card SaaS and business expenses not captured by infra cost collectors Runs monday 07:00.
   11. Fundraising Readiness Digest (fundraising-readiness): Weekly L3. Reads weekly-wbr scorecard + cost-collection daily files. Surfaces pre-seed story signals: MRR growth, signup acceleration, mission depth, capital efficiency, 7d return. No new collection, no LLM. Phase 4.2 G3. Runs monday 10:00.
   12. Growth Funnel Synthesis (growth-funnel): Weekly L3 synthesis. Reads weekly-wbr scorecard, reorganizes into a 4-stage funnel view (visitors → signups → missions → paid). Highlights tightest stage and biggest WoW shift. No new collection, no LLM. Phase 4.2 G2. Runs monday 09:00.
   13. Knowledge Discovery (knowledge-discovery): Identifies knowledge gaps, repeated patterns, and potential skills from fleet activity and regression analysis Runs monday 04:00.
   14. Lawyer Agent (lawyer-agent): Monthly legal risk assessment, scans internal commits and external regulatory developments, maintains company legal profile and risk register, produces lawyer-ready brief for EU/US/UK (GDPR, COPPA, FERPA, EU AI Act, Online Safety Act) Runs Monday 10:00.
   15. Link Rot Monitor (link-rot): Detects broken external links in MDX content files. Uses Sonnet only when broken links found. Runs wednesday 03:00.
   16. Outbound Attribution Digest (outbound-digest): Weekly outreach attribution analysis, which campaigns drive signups Runs friday 14:00.
   17. SEO Momentum Digest (seo-momentum): Weekly L3 synthesis digest. Reads from crawl-health, indexing-monitor, seo-regression to produce a single SEO momentum view: indexing snapshot, week-over-week deltas, top regressions, structural alerts. No new data collection, no LLM. Phase 4.2 G1. Runs monday 08:00.
   18. Verification Audit (verification-audit): Weekly audit of hallucination gate: coverage, expiring claims, bypass rates, PENDING_HUMAN backlog. Also triggers enforcement activation when calibration period ends. Runs Sunday 10:00.
   19. Weekly Business Review (weekly-wbr): Friday strategic analysis, metrics, trends, competitive landscape Runs friday 11:00.
   20. Weekly Roundup Compiler (weekly-roundup): Friday consolidated email, WBR + CFO + scouts + outbound + fleet Runs friday 22:00.
   21. Wiki Linter (wiki-linter): Weekly quality pass: finds contradictions, stale entries, orphans, and missing connections in the wiki Runs sunday 03:00.
7. Monthly, selected dates
   1. SEO Intent Attribution (seo-intent): Joins GSC clicks to PostHog sessions by intent, tracks conversion rates per intent bucket over time. Uses rules + Gemini Flash for query classification. Runs selected calendar dates.
8. Monthly
   1. Agent Manager (agent-manager): Monthly meta-agent, fleet review, prompt audit, state of art research, recommendations Runs monthly.
   2. Curriculum Scout (curriculum-scout): Monthly scan for curriculum changes across 18 target markets Runs monthly.
9. Retired
   1. Fleet Liveness Watchdog (RETIRED) (fleet-liveness-watchdog): RETIRED 2026-04-26. Job absorbed into fleet-health (which now does silence detection + stuck-process detection + SLA-schedule alignment). LaunchAgent disabled at ~/Library/LaunchAgents/com.flipeducation.fleet-liveness-watchdog.plist.disabled. Registry entry kept for historical reference. Runs retired.