ibrahimkobeissy/ai-second-brain-template: 🧠 AI Second Brain — an agent-operated PKM template for Obsidian

A self-pruning, high-signal second brain for Obsidian, operated by AI coding agents (Claude Code & Codex). Capture an idea → research it → synthesize a plan → ship actionable tasks. The vault is for thinking; the agents do the labor.

This repository is a template: clone it, open it in Obsidian, point your AI agent at it, and you have a working knowledge operating system — an ARA-structured vault plus a toolbelt of skills that turn raw bookmarks into strategic plans and tracked work. No personal areas or notes are included — you bring your own; the scaffolding and tooling are the product. The one deliberate exception is a single worked example — the standup-tools Area — shipped on purpose as an end-to-end demo of the Productize workflow (below).

✨ Why this exists

Most "second brain" setups rot: you capture endlessly and synthesize never. This template makes AI agents do the processing — curating captures, synthesizing drafts into plans, extracting tasks, and linting the knowledge graph — while you keep the judgment. It is agent-augmented, not agent-led.

• 🔁 A capture → curate → synthesize → act loop that prunes spent material as it goes (lite over large).
• 🧩 A toolbelt of skills (slash commands) that automate the boring parts.
• 🗂️ ARA structure (Areas / Resources / Archive) with a hard wall between Work, Personal, and Resources to stop context bleed.
• 🔗 Link-first: value lives in the [[wikilinks]], surfaced on a live dashboard.
• 🗣️ Calibrated to you: a one-time calibrate-comms test teaches the agents how you want to be written to, so output never lands as "too long" or "too terse".
• 🛡️ Security guardrails that stop agents from reading or exfiltrating your secrets (SSH keys, .env, cloud credentials, shell history).
• 🤝 Dual-agent peer review: one agent builds a tool, the other challenges and hardens it before it ships.

🚀 Quickstart

1. Use this template (or clone the repo).
2. Open the vault/ folder in Obsidian and install the community plugins it expects (listed in vault/.obsidian/community-plugins.json): Dataview, Tasks, Kanban, Chronos Timeline, Mermaid Tools, Advanced Tables, Editing Toolbar. Then turn on DataviewJS in Dataview's settings — the live dashboards (dashboard.md, per-Area dashboards) and the areas-activity-timeline.md need it. (Plugins aren't bundled — Obsidian installs them for you.)
3. Install Claude Code (and/or Codex) and open the repo as your project — the agent reads CLAUDE.md automatically (Codex reads it via the AGENTS.md symlink).
4. Calibrate how the agents write to you: /calibrate-comms — a 5-minute test so every doc afterward matches your style.
5. Create your first Area: /init-area.
6. Drop a link into vault/00-inbox/, then run /curate-bookmarks → /synthesize-drafts → /plan-to-kanban.

🔁 The Operating Loop

flowchart LR
    Inbox["00-inbox<br/>(raw bookmarks)"] --> Curate["curate-bookmarks"]
    Curate -->|extract signal| Draft["draft/<br/>(curated notes)"]
    Draft --> Synthesize["synthesize-drafts"]
    Synthesize -->|thematic synthesis| Plan["synthesis/<br/>(strategic plan)"]
    Plan --> Extract["plan-to-kanban"]
    Extract -->|dedupe + append| Kanban["todo-kanban.md<br/>(tasks)"]
    Kanban --> Execute["build / execute"]
    Execute --> Prune["delete spent<br/>intermediates"]

🗂️ Vault Structure (ARA)

🧰 The Toolbelt (Skills)

Run these as slash commands in Claude Code (or via the Skill tool); Codex invokes the same skills with $skill:

• /calibrate-comms — run this first. Profile how you want the agents to write to you — length, structure, visuals, jargon, tone — and persist those directives to CLAUDE.md so every agent matches your brain. Calibration, not personality typing.
• /init-area — scaffold a new Area: challenge the idea, define goals/scope, create the hub note, Kanban board, and triage queue.
• /curate-bookmarks — turn inbox captures into per-Area "what can we steal" drafts; judge each link independently; log the source.
• /scout-idea — challenge an idea's value, then run a broad discovery sweep for tools/OSS/articles when you have no bookmarks yet.
• /synthesize-drafts — analyze an Area's drafts against each other with a scientific thematic matrix and produce one strategic plan.
• /plan-to-kanban — extract the plan's action items into the Area's Kanban board, deduplicated.
• /productize-new → /productize-report — a six-phase product toolkit: take an Area idea from intake → PRD → market/competitive/feasibility analyses → an honest Go/No-Go → build specs + roadmap → a capstone visual HTML report. Honest-by-design (thin evidence → low confidence; a weak case earns NO-GO/PIVOT).
• /job-hunt-cv-from-doc → /job-hunt-starter — a personal job-hunt toolset built on career-ops as the execution hub: turn a CV into a data spine, grow a STAR story bank, prep & drill interviews, scout companies into HTML dossiers, and track progress — the skills feed career-ops and beautify what it can't. Ships as skills + an empty scaffold; your data stays private.
• /vault-linter — read-only graph integrity check: broken [[wikilinks]], orphan notes, missing traceability.
• /audit-maintenance — headless cross-agent peer review that challenges and hardens newly built tools.
• /security-guardrails — install a portable deny-list plus a PreToolUse hook that blocks shell reads of secrets, in any project.

🏭 Productize ��� take an Area from idea to build specs

Once an Area's thinking has matured, the productize toolkit runs it through a six-phase pipeline — from a one-line spark to a Go/No-Go decision and, if it survives, build-ready specs:

productize-new (intake → PRD) → productize-analyze (16 analyses) → productize-decide (Go/No-Go) → productize-build (deliverables + roadmap) → productize-report (capstone HTML).

It is honest by design: thin evidence yields low confidence, and a weak case earns a NO-GO or PIVOT — not a rubber stamp.

See it run end-to-end on a real, deliberately ordinary idea — StandupZero ("auto-draft each developer's daily standup from their real activity"):

• 📖 Walkthrough: Productize showcase — the StandupZero run, narrated phase by phase.
• 📂 Every artifact: the standup-tools Area — PRD, 16 analyses, Go/No-Go, 13 deliverables + roadmap, plus the visual summary report.
• 📘 Reference: Productize documentation — how the phases, control plan, and analysis dependency map work.

💼 Job Hunt — run a real search on a toolset you control

A personal job-hunt toolset (job-hunt-* skills) built on career-ops as the execution hub — the model is build the bespoke, adopt the commodity. career-ops does the heavy engine work (scan job boards, A–F deep-score each role, generate the ATS CV PDF, cover letters, outreach, application tracking); the vault skills feed it (turn your CV into the inputs it needs) and beautify what it can't (a shareable HTML portfolio, company dossiers, a progress dashboard).

Everything orbits one spine — your cv.md — which every skill (and career-ops, via a symlink) reads:

cv-from-doc (CV PDF/Word → structured cv.md) → craft-profile (portfolio HTML) · capture-story (STAR story bank) · interview-prep · mock-interview → careerops-profile / careerops-portals (configure career-ops) → career-ops scan + A–F eval → scout-company (dossier + me-vs-the-JD matrix) → apply via career-ops → interview-debrief (mines new stories back into the bank). Run job-hunt-starter anytime for an on-disk progress dashboard that nudges you to the next step.

Privacy by design: unlike the StandupZero demo, this ships as skills + an empty folder scaffold only — your CV, stories, and dossiers live under hunter/ and never leave your own private remote. Nothing personal is committed to this template.

• 📖 How to run it, step by step: Job-Hunt toolset manual — the full journey, the vault ↔ career-ops switching points, and the flywheel that makes every interview sharpen the next.
• 📂 The scaffold to fill: the job-hunt Area — the ready-to-use structure (hunter/, draft/, scout/, synthesis/) you drop your own CV into.

📚 Documentation

• Operating System doc — the full philosophy, the loop, peer review, and the complete skill reference.
• Calibrate Comms — the one-time test that tunes how the agents write to you (run it first).
• Productize toolkit — the six-phase idea → build-specs pipeline, with the StandupZero worked example.
• Job-Hunt toolset — the job-hunt-* skills + career-ops workflow: CV spine, story bank, interview prep, company dossiers, and progress dashboard.
• Vault conventions — frontmatter schema, tags, titles, link-before-close.
• CLAUDE.md — the master instruction set both agents follow (AGENTS.md is a symlink to it, so Codex reads the same rules).

🛡️ Security

AI agents operate inside the project directory only. The included security-guardrails skill wires up a permissions.deny list and a PreToolUse hook so agents can't read or exfiltrate credential stores, .env files, private keys, tokens, or shell history — even via a shell cat. Harden any repo with /security-guardrails.

🤝 Contributing

Issues and PRs welcome — especially new skills and workflow improvements. New tooling is expected to pass the dual-agent peer review described in the Operating System doc.

📄 License

CC BY-NC-SA 4.0 © 2026 Ibrahim Kobeissy. See LICENSE.

You may use, adapt, and share this template for non-commercial purposes with attribution, and must license any derivatives under the same terms. Commercial use requires separate permission.

This template does not bundle third-party Obsidian plugins. Only vault/.obsidian/community-plugins.json (the list of plugins to enable) ships; you install the plugins yourself via Obsidian (see Quickstart), so each stays under its own author's license.