Open-source index
Every public repo behind these docs, one line each, with the article that explains it.
Every public repo that relates to these docs, one line each, with the article that explains it. All of them live under github.com/jddavenportOpen, and each one was checked with a logged-out request on October 5, 2026. My private system stays private because it runs my actual life and work. These are the parts that generalize.
Most are MIT licensed; check each repo's license file before you reuse code. Where a repo's own README quotes an agent count, it was true when that README was written; the durable phrasing is 30+ agents in daily production, 100+ built over time.
Production core
The safety, governance and orchestration pieces, extracted from the system that runs every day.
| Repo | What it is | Read |
|---|---|---|
| warden | A governor for autonomous agents: autonomy admitted by reversibility, not confidence. Zero dependencies, offline tests | Auto bug-squash and the Warden |
| claude-bug-squash | Point Claude Code at a repo and a bug: it fixes in an isolated worktree, proves red to green, gates the fix, and never merges out of the box | Auto bug-squash and the Warden |
| ethos-gate | PreToolUse and Stop hooks that block irreversible or outward-facing actions unless a human approved them | Hooks |
| claude-deploy-kit | Enterprise controls for Claude agents: a policy chokepoint, a destructive-command deny hook, an eval gate, a redacting audit log, pluggable secrets | One door out: the send gate |
| orchestra-agents | A small Python library for driving many Claude agents in parallel with guardrails and evals: fan-out, a durable bus, circuit breakers | Orchestration: delegate down, verify up |
| clawd-agent-os | Tiered agent hierarchies with worktree-isolated fleets, a durable agent bus and one model registry as the source of truth | Fleets: parallel build agents |
| hub-watchdog | A two-strike uptime watchdog that runs on GitHub Actions, so it keeps working when the machine it watches is down | Watchdogs |
| agent-safety-case-study | A capability-versus-risk case study of running an AI organization in production: each capability, its risk, the control, an incident | The architecture in one page |
| claude-code-behavior-specs | Target-behavior specs for the Claude Code CLI written from production incidents, each with a reproduction and the workaround that shipped | Lessons from running agents every day |
| nerve-center-showcase | The capabilities page my system regenerates from live state every day (live page) | The system documents itself |
Agent tooling
Things you can drop into your own Claude Code setup.
| Repo | What it is | Read |
|---|---|---|
| agent-evals-starter | Learn agent evals by running them: a support agent, 24 tasks, code graders, an LLM judge, controls. Runs offline | Agent evals |
| mcp-judge | A calibrated multi-persona LLM judge as an MCP server, with an eval harness that proves the calibration | How to validate your agent eval |
| context-kit | Four personal-context templates plus five Claude Code skills, so an agent starts with real context about you. One-command install | Personal context files |
| harnessview | Scan a project's Claude Code setup and see it as one live graph that flags broken wiring | harnessview |
| deep-research-agent | Iterative, multi-source research with a knowledge-gap loop, a devil's-advocate pass and per-claim verification | The deep-research agent |
| mastery-engine | Point it at a codebase and get a curriculum about your actual code, with every claim verified against disk | No article yet |
Personal AI OS
| Repo | What it is | Read |
|---|---|---|
| clawdling | The open, bring-your-own-key personal AI OS engine: chat that acts, a multi-pane Claude Code cockpit, headless workers. AGPL-3.0 | Clawdling: the open engine |
| fleetwright | A self-hosted agent operations OS: a persistent chief of staff that commands a fleet of agents | The CEO agent |
| openplaud | A free, self-hosted manager for Plaud voice recordings, with your own transcription keys or local Whisper | Voice notes into the second brain |
Apps
| Repo | What it is | Read |
|---|---|---|
| resume-grader | One Claude call scores any resume against any job on a six-dimension rubric | Build your own resume grader |
| vc-job-board | The job board behind the BYU AI Foundry jobs page: venture portfolio roles, twice-daily ingest, resume matching | The BYU AI Foundry |
| recruit-copilot | A job search as a verification problem: one experience bank, tailored resumes proven machine-readable, no auto-apply | recruit-copilot |
| openbudget | Self-hosted envelope budgeting with read-only bank sync through SimpleFIN | OpenBudget |
| voiceclaw | A voice agent you call by phone, on Vapi, Claude and Deepgram. Archived | VoiceClaw: a phone agent |
| bookjd | A self-hosted booking app, your own Calendly: real calendar availability, optional payment, both calendars updated | No article yet |
| jddavenport-site | Source for jddavenport.com | About these docs |
Not listed here
A few public repos are older experiments or side projects that these docs don't cover. Browse the full list on GitHub if you're curious. Anything not on that page isn't public, whatever an older page may have linked.
Next: About these docs