feat: add Claude Code skills and expand agent documentation

[Jazzcort]

Motivation

The project lacked structured guidance for Claude Code agents — there was no CLAUDE.md, the AGENTS.md was a minimal stub, and there were no reusable skills for common contributor
workflows (committing, creating PRs, reviewing code). Contributors and agents had to rediscover project conventions each time.

Summary

• Added three Claude Code skills (generate-commit-message, prepare-for-pull-request, review-pr) to standardize contributor workflows
• Created CLAUDE.md with core operating rules and a pointer to AGENTS.md
• Expanded AGENTS.md from a brief stub into a comprehensive agent reference covering architecture, project layout, tools, configuration, tech stack, and security

Changes

Docs:

• New CLAUDE.md — four core rules for agent behavior plus project details reference
• Rewrote AGENTS.md — added architecture diagram, detailed project layout, tools overview table (fixed + guarded), configuration reference for all LINUX_MCP_* env vars, tech stack
  section, and expanded security guidance (SSH key-only auth, host key verification, audit log redaction)
• Added security docs pointer (docs/security.md) to the Docs section

Claude Code Skills:

• .claude/skills/generate-commit-message/SKILL.md — analyzes staged changes and generates conventional commit messages using project-specific type/scope conventions
• .claude/skills/prepare-for-pull-request/SKILL.md — runs make verify, gathers changes since a base ref, and generates a structured PR description
• .claude/skills/review-pr/SKILL.md — six-dimension PR review (format, types, semantics, readability, redundancy, bugs/security) with verdict criteria

🤖 Generated with Claude Code

---

Just some experiments with skills. Feel free to try it out! we can merge this if it really helps.

Usage:

• /generate-commit-message:
  Stage your changes, then invoke the skill

  git add src/linux_mcp_server/tools/network.py tests/tools/test_network.py

  Invoke /generate-commit-message in claude code

  It inspects git diff --cached, checks recent commit style, and produces a conventional commit message like:
  feat(network): add protocol filter to get_listening_ports

• /prepare-for-pull-request:

  Invoke it with a base ref (or let it auto-detect from reflog):

  /prepare-for-pull-request main

  It runs make verify first (lint, types, tests). If everything passes, it analyzes all commits since main, then generates a structured PR description with Motivation, Summary, Changes,
  and Test plan sections. You can then create the PR directly or edit first.

  Without arguments it auto-detects the branch point:

  /prepare-for-pull-request

• /review-pr:

  Pass a GitHub PR URL:

  /review-pr https://github.com/yourusername/linux-mcp-server/pull/42

  It fetches the PR, checks out the branch, and runs a 6-dimension review: format/lint, type check, semantic correctness, readability/redundancy, bugs/security, and tests. The output is a
  structured report ending with a verdict (APPROVE, REQUEST CHANGES, or NEEDS DISCUSSION) and actionable findings with severity levels.

[github-actions]

For team members: test commit af71bee in internal GitLab

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Flag	Coverage Δ
unittests	96.54% <ø> (-0.38%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.
see 9 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[github-actions]

For team members: test commit bedb237 in internal GitLab