SpecSelf Starter

Quick start

From the button to your first session.

Three steps. Copy-paste ready. Under five minutes of reading, then the agent does the interviewing.

The three steps

  1. Get the template

    Two main paths, plus a way to browse first. The button below routes through our counted redirect straight to GitHub's “Use this template” page.

    Use this template

    Free and MIT-licensed. No account, plain markdown — the button takes you to GitHub.

    We count this click server-side and measure page views with Google Analytics. The kit itself ships with no tracking.

    • Use this template on GitHub (recommended) — creates your own private repo.
    • Want to read the README first? Browse the repo before you decide.
    • Want to see a finished corpus first? Browse the living example.
    • Or clone it for the local-only setup:
    git clone https://github.com/CaskeyCoding/specself-starter.git

  2. Point your agent at it

    Open the repo with your agent and paste this exact line:

    Read `AGENTS.md` and everything in `steering/`, then walk me through `GETTING-STARTED.md`.

    Any capable agent works via AGENTS.md. With Claude Code you also get the .claude/ extras — persona subagents, the cadence skills (/foundation, /daily, /weekly, /monthly, /quarterly, /annual), the /capture, /coherence-check helpers, and an overdue-review nudge.

  3. Run the first foundation sitting

    Once, top of the stack down, but not in one block: a first 15-to-20-minute sitting drafts your opening principles, and the rest spreads across three more short sittings over week one, so the biggest ask never lands on day one. The agent interviews you; you author. The five passes it works through:

    • Principles — from stories, not values; rules you can back with evidence.
    • Vision — one present-tense page of who you're becoming.
    • Areas — the standing roles you maintain, each with a standard.
    • Goals — finite outcomes with a falsifiable success criterion, each serving something above it.
    • Projects — what you're driving now, plus one to three habits.

    The agent writes only candidates you edit; it authors nothing as committed. Already arriving with years of notes? Run /adopt to draft a first corpus from them instead of authoring from scratch. Full script: GETTING-STARTED.md.

The one rule

You decide and commit. The agent organizes, connects, surfaces, and challenges — but never silently authors your life.

Then what? It becomes a loop: /capture or /daily in a couple of minutes, /weekly to cluster and coherence-check, and the deeper /monthly / /quarterly / /annual reviews to synthesize and prune. The diff across time is the insight.

Your content stays yours

Your corpus lives under your-life/, gitignored by default — commit the structure, keep the content private. Nothing is sent to us; there is no backend. Choose an agent provider whose data policy you accept, since the model sees whatever you put in a session.

This site uses Google Analytics to count visits. Nothing in the kit or your corpus is tracked.

Read the privacy model →