Prometheus
Personal Claude Code configuration — 32 skills, 20 hooks, cross-device sync
A Claude Code configuration repo that clones into ~/.claude and turns the
vanilla CLI into an opinionated assistant with memory, specialized skills,
voice output, event capture, and auto-sync. Not a product. Not a demo. The
assistant I actually run.
What it is
Prometheus is personal infrastructure, not a shippable app. One git repo
lives at ~/.claude on every machine I use (two Macs + Termux on Android).
A SessionStart hook pulls the latest; a SessionEnd hook pushes back.
The "brain" — skills, hooks, learnings, profile, knowledge — syncs via
git. Raw session data, vector stores, and device-specific binaries stay
local. Inspired by Daniel Miessler's PAI and its kai CLI, built around
my own skill set, life state, and the pillar framing I use (ERGON /
LOGOS / ARETE / OIKOS / TELOS).
By the numbers
| Metric | Value |
|---|---|
| Custom skills | 32 |
| Lifecycle hooks | 20 |
| Sync target | ~/.claude on every device |
| Devices in lockstep | 3 (two Macs + Termux on Android) |
| Commits in build window | 16 between 2026-01-03 and 2026-01-05 |
| Status | Dormant repo, still the live ~/.claude |
| Portable | No — hardcoded paths, pinned Node, macOS-only voice |
| License | Private — contains personal profile + life state |
Architecture
Device A (Mac) Device B (Mac) Device C (Termux)
| | |
+-- ~/.claude/ <---- git <---- ~/.claude/ ---+
|
v
SessionStart -> pull-prometheus.ts
SessionEnd -> sync-prometheus.ts
|
v
Synced: skills/ hooks/ agents/ history/
knowledge/ config/ profile + life-state
Local: memory/.venv raw session data device binsEvery Claude Code event routes through a hook: SessionStart,
PreToolUse, PostToolUse, UserPromptSubmit, Stop, SubagentStop,
SessionEnd. The prometheus CLI at bin/prometheus wraps Claude Code
with context load, voice, and skill dispatch in a single command.
Key features
- 32 skills under
skills/— each with its ownSKILL.mddefining triggers, workflows, and domain knowledge. Coverage: software dev (Coder, SoftwareDev, Architect, DevOps, QA, Security, Data), research (Research, Journalist, SEC, FEC, FederalRegister, Census, FBICrime, Congress), writing (Writing, WebEssay), meta (CORE, Orchestrator, Agents, CreateSkill, SelfOptimize, SelfUpdate, Prompting), ops (Git, NotionSync, WebScraper, Seed, CreateTests, SpecKit), and Art. - 20 lifecycle hooks under
hooks/— auto-pull on start, core context load, proactive suggestions, security validation, tab-title updates, voice confirmations, session summarization, learning extraction, auto-push on end. - Cross-device sync —
pull-prometheus.tsat session start,sync-prometheus.tsat session end. Verified working on MacBook Pro- Termux.
- Vector memory at
memory/—sentence-transformersfor semantic search across extracted learnings. Per-device.venv, intentionally not synced. - Knowledge pipeline —
bin/fetch-ai-sources.pypulls from a curatedAI.json(YouTube channels, repos, sites); Claude synthesizes intoknowledge/synthesized/*.md. - Agent personality v0.9 (tag
JARVIS v0.9) — proactive action hooks, agent voice mapping, observability layer. - Single CLI entry point —
prometheus "..."wraps Claude Code with context, voice, and dispatch. Inspired by Miessler'skai.
What makes it stand out
- Honest about what it is — a config bundle, not a product. The repo is dormant because the patterns graduated into daily Claude Code usage; it stopped needing commits.
- Git as sync transport — no bespoke backend. The repo IS the sync layer. Personal data is in-repo because the repo is private.
- Opinionated, not portable — hardcoded paths, pinned Node
version (
~/.nvm/versions/node/v20.19.6), macOS-specific voice (say). Anyone else cloning it rewritessettings.jsonfirst. - Philosophical framing baked in —
fardin-profile.mdandfardin-life-state.mdload at session start. Pillars and goal (eudaimonia, not comfort) are context, not aesthetics.
Stack
| Layer | Technology |
|---|---|
| Runtime shell | Claude Code (hooks, skills, settings.json) |
| CLI + hooks | Bun + TypeScript, tsx via nvm-pinned Node 20.19.6 |
| Vector memory | Python 3 + sentence-transformers + torch |
| Voice | macOS say + ElevenLabs |
| Integrations | Notion API, Linear API, Google APIs |
| Sync transport | Git + GitHub |