gitclaw agent · linux kernel

Your Kernel Mentor
on demand

Senior kernel engineer + incident commander in one agent. Paste a panic, ask about internals, debug live — free, no login.

kernel-mentor-agent — gitclaw run .
$ gitclaw run .
✔ agent: kernel-mentor-agent v1.0.0
✔ skills: explain-kernel, debug-code, setup-env
✔ model: claude-sonnet-4-5 · temp: 0.2
 
agent> why is my kernel panicking in mm/slub.c?
 
⬡ Kernel Failure Radar
Subsystem: mm / slab allocator
Hypothesis 1 [HIGH 78%]: use-after-free in kmem_cache_alloc
Hypothesis 2 [MED 54%]: double-free via corrupt slab pointer
Next probe: kasan_check_read / KASAN report · L1
 
agent>
// gitclaw

Run locally with gitclaw

Three commands and the full agent runs in your terminal with your own model. No cloud, no signup.

01
Install gitclaw
gitclaw is a CLI tool that reads agent.yaml and runs agents locally. Install it globally via npm.
npm install -g gitclaw
02
Clone this repository
Get the agent package with all skills, SOUL, and RULES onto your machine.
git clone https://github.com/sachindra535/kernel-agent.git
cd kernel-agent
03
Run the agent
gitclaw picks up agent.yaml automatically. The agent starts and you can chat right in your terminal.
gitclaw run .
04
Try a prompt
The agent is running. Paste a kernel panic, ask about internals, or set up your dev environment.
agent> Debug this kernel panic: Unable to handle kernel NULL pointer dereference at 0000000000000010
Open source · MIT license
sachindra535/kernel-agent
Skills, SOUL, RULES, and full agent.yaml — all in one repo. Fork and customize.
↗ View on GitHub
// capabilities

What the agent knows

Three skills. Two signature features. Built for engineers shipping hard kernel work under pressure.

Kernel Failure Radar
Maps sparse signals — panic text, dmesg, perf anomalies — to subsystems with confidence-ranked hypotheses and the three highest-yield next commands.
Patch Safety Ladder
Labels every change recommendation L0–L4 so you move fast without reckless breakage. Rollback plan required before L4 rollout.
explain-kernel
Layered internals explanations: mental model → kernel call path → key structs and APIs → verification commands. Teaches while solving.
debug-code
Incident-grade debug workflow. Non-destructive probes first. Every error output is explained and mapped to a concrete next action.
setup-env
Profile-based environment bootstrap: fast-lab, repro-lab, incident-lab. Readiness scorecards and rollback-validated boot paths.
gitclaw native
Full gitagent layout. gitclaw run . with 20-turn context, 0.2 temperature. Works with any Anthropic-compatible model.