agent-tooling

Three small tools for running, watching, and searching fleets of Claude Code and Codex sessions. They were built for one person's day-to-day workflow on WSL + Windows Terminal, and they're shared warts and all — expect hardcoded assumptions, WSL-isms, and rough edges. Nothing here is a product; it's a working setup you're welcome to lift, fork, or cannibalise.

                         ┌───────────────────────────┐
   ~/.claude/projects ──▶│      santa  (search)      │  semantic search + resume
   ~/.codex/sessions  ──▶│  indexes every transcript │  over your whole history
                         └────────────┬──────────────┘
                                      │ "resume this session"
                                      ▼
                         ┌───────────────────────────┐
                         │     cockpit  (control)    │  many live sessions in one
                         │  tmux grid of live panes  │  colour-coded tmux grid
                         └────────────┬──────────────┘
                                      │ "open N agents as a workspace"
                                      ▼
                         ┌───────────────────────────┐
                         │ agent-fusion (multi-agent)│ Claude + Codex on one task,
                         │   blind-first, then fuse  │  output fused
                         └───────────────────────────┘

They're independently useful, but they compose: santa finds and resumes sessions, cockpit is the grid you resume them into, and agent-fusion launches multi-agent runs as cockpit workspaces.

The three repos

Repo	Language	One-liner
santa	C# / .NET 10	Local semantic search + browse + resume over all your Claude & Codex transcripts. Embeddings + reranker run fully on-device.
cockpit	Bash + tmux	A control surface that resumes several unfinished sessions into one titled, colour-coded, live-state tmux grid you steer from the keyboard.
agent-fusion	Bash + a protocol doc	Runs 2–3 independent agents (Claude/Opus + Codex/GPT) on one task and fuses their output. The filesystem is the coordination bus.

Each repo has its own README with the real detail. This one is just the map.

Quick start

You don't need all three. Pick what's useful. Rough order of dependencies: santa stands alone → cockpit is best with santa → agent-fusion wants cockpit + the claude/codex CLIs.

# 1. santa — search & resume your history  (needs the .NET 10 SDK)
git clone https://github.com/OvenBaker/santa && cd santa
./install.sh && santa tui     # full-screen browse + search; or one-shot: santa query "…"

# 2. cockpit — drive many sessions at once (needs tmux ≥3.4, jq, sqlite3)
git clone https://github.com/OvenBaker/cockpit
export PATH="$PWD/cockpit:$PATH" && cockpit

# 3. agent-fusion — multi-agent runs
git clone https://github.com/OvenBaker/agent-fusion && cd agent-fusion
./fusion --mode A <slug> task.md --dry-run    # preview prompts, spawn nothing

Dependencies (consolidated)

Need	For	Notes
.NET 10 SDK	santa build	install.sh does a single-file dotnet publish.
bash, tmux ≥3.4, jq, sqlite3	cockpit	Core runtime.
claude CLI	cockpit, agent-fusion	Claude Code.
codex CLI	cockpit, agent-fusion (optional)	Only if you want Codex sessions tracked / fused.
WSL + Windows Terminal	cockpit (assumed)	The shim/wt.exe trick is WSL-specific; see the warts below.
Internet, first run only	santa	Pulls embed/rerank models (HuggingFace) + the sqlite-vec extension (GitHub). All on-device after that.

Warts & assumptions (read before you judge it)

• WSL-centric. cockpit assumes WSL + Windows Terminal; shim/wt.exe is a stand-in so santa's resume can hand a session to cockpit. On bare Linux the grid still works, but the wt.exe integration and a couple of paths will need adjusting.
• agent-fusion is coupled to cockpit for launching only. The protocol (agent-fusion.md) and the coordination lib (fusion-lib.sh) are pure filesystem and tmux-agnostic. --dry-run prints each agent's prompt so you can run a fusion by hand in any N terminals, no cockpit required.
• agent-fusion's launcher pre-trusts the run dir. It writes trust/hasTrustDialogAccepted records into ~/.codex/config.toml and ~/.claude.json so fresh run dirs don't stall on the CLI's "do you trust this folder?" prompt. Convenient, but it's a security decision made on your behalf — read fusion before running it unattended.
• santa indexes everything you've ever said to Claude/Codex. The index DB is local-only and gitignored; just know it exists and where it lives (~/.local/share/santa/).
• GPU is optional. santa ships CUDA onnxruntime providers but falls back to CPU. Embedding a large history on CPU is slower but fine.
• Hardcoded niceties. Default model IDs, the tmux -L cockpit socket name, the ~/agents fusion home (override with FUSION_DIR), ~/.local/share/santa (override with SANTA_HOME) — all editable, none parameterised behind a config UI. It's personal tooling.
• No tests, no CI. Treat it accordingly.

License

The Unlicense — released into the public domain. Do whatever you want.