The daily loop: four phases, fully automated

Morning: Intelligence Briefing and Memory Load

The day starts at 9 AM with two automated routines. First, the PR briefing skill scans my GitHub repos for overnight commits, open pull requests, and new issues. It posts a plain-English summary to Discord so I can see what changed while I was offline.

Second, the session startup reads persistent memory. That means loading MEMORY.md (curated facts and preferences that carry across every session), yesterday's daily log, and the project Kanban board. If I captured notes from my phone overnight via the Discord bot, those get fetched from Google Drive, sanitized for prompt injection, and surfaced in the first prompt.

The result: I sit down and the system already knows what I was working on, what is in progress, and what I noted on my phone at 2 AM.

Midday: Autonomous Content Pipeline

The daily AI news post runs through a fixed pipeline. NotebookLM researches the top stories, generates an infographic and briefing report, and I write a structured post anchored to the infographic's data points. Pre-deploy checks run automatically (image size limits, slug validation, Playwright page verification), and then it goes to chento.io and Buttondown subscribers.

I do not format anything manually. The Astro template, SEO metadata, structured data, and social preview images are all generated from the same source material. The content calendar tracks topics through five stages (planned, researched, drafted, published, distributed) in a SQLite database.

44 skills handle everything from blog posts to security scans to meeting prep. 11 agents coordinate the more complex workflows. The content pipeline alone uses four of them.

Continuous Security and Self-Healing

The self-healing agent runs twice daily. It audits code quality, hardens scripts, checks skill health, and verifies the live site is responding correctly. Reports get archived to the Obsidian vault.

Five Docker-based security scanners (Nessus, OWASP ZAP, Semgrep, Trivy, Gitleaks) run vulnerability assessments. Results post to Discord and feed into a local SOC dashboard that visualizes posture, attack surface, and findings over time.

Every piece of external content that enters the system, including mobile notes, API responses, and memory search results, passes through a prompt injection sanitizer before it touches Claude's context. That is non-negotiable.

Evening: Session Wrap and Vault Sync

At the end of a session, notable events get logged to the daily note. Both repos (main codebase and Obsidian vault) get committed and pushed. The Obsidian Git plugin auto-syncs the vault to a private GitHub repo, so everything is backed up and accessible from any device.

The Kanban board in ROADMAP.md gets updated: items move from In Progress to Completed, new backlog items get added, and tomorrow's priorities get surfaced.

The system is not perfect. GCP auth tokens expire in worktrees. NotebookLM infographics sometimes get the color palette wrong. Scripts break and need fixing. But the structure holds, and the structure is the whole point. One person, 44 skills, and an AI that remembers what happened yesterday.

Originally published on chento.io