Android Kernel LKM (SKILL.md)

Standard Agent Skills definition for developing Android kernel loadable modules (KO/LKM) with kprobe/kretprobe.

Compatible with: OpenCode, Claude Code, and Codex (Agent Skills standard).

Install

The skill folder contains references/ and scripts/ alongside SKILL.md. Copy the whole folder, not just SKILL.md — the entry point links to those files, and scripts/extract-abi.py is part of the baseline-extraction step.

Any Agent Skills-compatible tool uses the same layout: put the whole folder under that tool’s skills directory.

OpenCode

git clone https://github.com/Yunnijian/android-kernel-lkm \
  ~/.config/opencode/skills/android-kernel-lkm

Or copy an existing checkout: cp -R android-kernel-lkm ~/.config/opencode/skills/.

Claude Code

git clone https://github.com/Yunnijian/android-kernel-lkm \
  ~/.claude/skills/android-kernel-lkm

Codex

git clone https://github.com/Yunnijian/android-kernel-lkm \
  "${CODEX_HOME:-$HOME/.codex}/skills/android-kernel-lkm"

Verify

The tool should list android-kernel-lkm as an available skill after installation. The helper script can be checked directly:

python3 /android-kernel-lkm/scripts/extract-abi.py --help

agents/openai.yaml sits inside the folder and is read from there; it is optional and Codex-only.

Usage

The skill triggers when the user mentions: KO, LKM, 内核模块, kernel module, kprobe, kretprobe, or insmod.

The skill guides the agent through:

• Phase 0: Feasibility research (GitHub + device reality)
• Phase 1: Requirements gathering
• Phase 2: Baseline extraction (vermagic, CRC, kheaders, KCFI)
• Phase 3: Symbol discovery
• Phase 4: Struct layout discovery
• Phase 5: Iterative development (minimal probe → progressive, full static verification + insmod testing per iteration)
• Phase 6: Final review (code + requirement fulfillment)
• Phase 7: Packaging and delivery (load-stage selection, boot guard, module zip + ko + hash)

Structure

android-kernel-lkm/
├── SKILL.md                        # Skill definition (Chinese + English)
├── agents/
│   └── openai.yaml                 # Codex-specific extension (optional)
├── references/                     # Read on demand
│   ├── abi-provenance.md           # Is the baseline the copy the device runs?
│   ├── boot-lifecycle-and-recovery.md # Load stage, boot guard, kernel gate, failure recovery
│   ├── display-mode-validation.md  # Enumeration vs timing vs actual refresh validation
│   ├── kcfi-and-dispatch.md        # KCFI limits, vendor dispatch guards, index shifts
│   └── rollback-and-forensics.md   # Rollback order, crash forensics, log capture
├── scripts/