Behavioral guidelines and engineering skills for AI coding agents. Research into optimal Soul/CLAUDE.md configurations for Claude Code and Hermes Agent.
.claude4me/
├── CLAUDE.md # Karpathy CLAUDE.md (原文摘录)
└── skills/
├── advanced-tdd/ # MattPocock: Vertical-slice TDD, behavior-over-implementation
├── deep-review/ # AddyOsmani: 5-axis code review with severity labels
└── diagnose-loop/ # MattPocock: Reproduce → Minimise → Hypothesise → Fix
| Skill | Source | Why |
|---|---|---|
| advanced-tdd | mattpocock/skills | Vertical slices, tracer bullets, behavior tests over implementation details |
| deep-review | addyosmani/agent-skills | 5-axis review (correctness/readability/architecture/security/performance), severity labels |
| diagnose-loop | mattpocock/skills | Disciplined debug loop: feedback loop first, 3-5 hypotheses, tagged instrumentation |
Not included (conflicts with native agent capabilities):
- Planning/task breakdown — breaks Hermes
subagent-driven-developmentand Claude Code's task flow - Git workflow — native GitHub skills are deeper and platform-specific
Option A: Plugin directory (recommended)
git clone https://github.com/<your-repo>/.claude4me.git /path/to/.claude4me
claude --plugin-dir /path/to/.claude4meClaude Code reads .claude-plugin/plugin.json and auto-loads all listed skills.
Option B: Copy skills into project .claude/skills/
# In your project root:
mkdir -p .claude/skills
cp -r /path/to/.claude4me/skills/* .claude/skills/Option C: Inline in CLAUDE.md
# Append SOUL + all skills to your project's CLAUDE.md
cat CLAUDE.md > /path/to/your-project/CLAUDE.md
for skill in skills/*/SKILL.md; do
echo -e "\n---\n" >> /path/to/your-project/CLAUDE.md
cat "$skill" >> /path/to/your-project/CLAUDE.md
doneSkills go in $HERMES_HOME/skills/ (default ~/.hermes/skills/). Each skill is a directory containing at least SKILL.md.
# Clone the repo
git clone https://github.com/<your-repo>/.claude4me.git /tmp/.claude4me
# Copy skills to Hermes skills directory
cp -r /tmp/.claude4me/skills/advanced-tdd ~/.hermes/skills/advanced-tdd
cp -r /tmp/.claude4me/skills/deep-review ~/.hermes/skills/deep-review
cp -r /tmp/.claude4me/skills/diagnose-loop ~/.hermes/skills/diagnose-loopOr use the Hermes CLI:
hermes skills list # Verify installed
hermes -s advanced-tdd # Preload a specific skill for a sessionTo load a skill mid-session, use /skill <name>.
⚠️ Do NOT usehermes skills install— the hub registry only downloadsSKILL.md, not full directories with supporting files. These skills include reference docs and scripts, so manual copy is required.
mkdir -p .cursor/rules
cp skills/advanced-tdd/SKILL.md .cursor/rules/advanced-tdd.md
cp skills/deep-review/SKILL.md .cursor/rules/deep-review.md
cp skills/diagnose-loop/SKILL.md .cursor/rules/diagnose-loop.mdCursor auto-loads rules from .cursor/rules/.
Copy the SKILL.md files into whatever skill/rules directory your agent uses. The format is standard markdown with YAML frontmatter — no agent-specific dependencies.
- Declarative boundaries over identity manuals — "what to do/what not to do" not "who I am"
- Platform-agnostic baseline — no Docker/WeChat specifics in SOUL
- Source-faithful — direct quotes from references, no AI rewriting
- Minimal — only skills that address real failure modes
- Karpathy CLAUDE.md: Simplicity First, Surgical Changes
- addyosmani/agent-skills: 5-axis review, change sizing
- mattpocock/skills: Vertical TDD, diagnose loop