Things I built that you can use

Instruments

• AI Usage Cardshipped · in use

  My own skill: a prompt you run in your own Claude Code that measures how you actually operate the model — your level (L0–L5), your ≈ Cherny adoption step, your one next unlock. 100% local, ~2 minutes. Live at silou.dev/pulse, and open-source in my skills repo; the story is in behavior over kit.

• cLensopen-source · MIT · beta

  A local-first, zero-dependency Claude Code observability instrument that turns invisible agent behavior into auditable signal. The transcript tells you what your agent typed — cLens shows you why: it captures and analyzes every session, and your code never leaves your machine. Terminal-first today, web front-end in progress.

  clens.dev has the real thing — what it surfaces, screenshots, and early pricing. It's still in beta, so the paid tier is a notify-me rather than a checkout; I'd rather take your email than your money before it's earned. Code: github.com/silouone/clens (npm @silou/clens).

• Software factoryprivate · building

  An autonomous dev-workflow system I'm building for my own work: typed tickets in, human-reviewed pull requests out — agents and deterministic code in the middle, humans at both ends. Spec-driven, and instrumented end-to-end by cLens + OpenTelemetry on every run. Private today; it's the working proof behind "run it without you" on the home page.

• Personal knowledge hubdaily use

  A memory system, not a note pile. Zettelkasten at its heart — atomic notes that earn their keep by what they link to — and spaced repetition as the layer that actually moves a notion into long-term memory, which is the highest-leverage part and the one most tools skip. Agents work over a local database served through MCP, and my ingest skill is the front door: point it at a talk or an article and it comes back as a key-points extraction assessed against my actual setup, ready to become notes. The skill is the only public piece — the hub itself is mine, not a product, and it's the testbed for patterns I later teach.

The harness, in the open

The machinery the instruments run inside — the agentic layer, the hooks, and the parts I didn't write. Same bar as everything above: what's shipped says shipped, what's half-built says so.

• Agent skillsopen-source · MIT

  The skills I author and use daily, published for both Claude Code and Codex — because half the people I share a skill with run a different runtime, and hand-copied ports drift within weeks. Each skill is one platform-neutral contract plus two maintained adapters, with a drift checker that fails the build when a core changes and its adapters weren't revisited. Currently ai-usage-card, ingest and sitrep: github.com/silouone/skills.

• Voice loopopen-source · MIT

  Talk to your coding agent; it talks back. When Claude Code or Codex finishes a turn, a hook turns the agent's final message into a ≤220-character spoken recap — computed deterministically, no model call and no extra tokens — and speaks it while you're across the room. The details that took real iteration: it detaches in ~0.1s so it never blocks a turn, a newest-wins debounce stops parallel sessions stacking audio, each agent and repo can have its own voice so you know who's talking, and it degrades to the offline macOS voice when the API is down. Paired with speech-to-text for the input half, you can run whole iterations without touching the keyboard. github.com/silouone/llm-voice-loop.

• What I run that isn't minecredits

  A harness is mostly other people's good ideas, wired together, and I'd rather say which is which. The agentic-engineering patterns underneath mine came from IndyDevDan; the LLM fundamentals from Matt Pocock's AI Hero — both still on my own reading list. The voice loop above speaks through Higgs Audio (Boson AI) and listens through SuperWhisper; I'm not affiliated with either, and neither is required — it falls back to what your machine already has.

Prefer the live version?

The clearest read on how I work is a session on your own code.