feat(install): Hermes and agy install paths

[ruslanlap]

What

The skill worked in Claude Code only. Two other runners want it, and neither needs code — the whole thing is a text ruleset.

agy / Antigravity CLI

agy has no --system-prompt flag:

flags provided but not defined: -system-prompt

So the ruleset has to ride in the prompt. Verified live against gemini-3.8-flash-low — the model answers in cavemenko style:

sed '1,/^---$/d' skills/cavemenko/SKILL.md > rules.md
agy -p "$(cat rules.md)

Поясни, чому useEffect з [] не викликає ре-рендер, а з [count] — викликає." \
    --model claude-opus-4-6-thinking

Project-level alternative: drop skills/cavemenko/SKILL.md in as AGENTS.md and agy loads it for every session. No agy setup, no wrapper daemon — it is a prompt file.

Hermes

The same SKILL.md installs as a Hermes skill, and is published at /opt/data/skills/productivity/cavemenko with an added When to Use section. Levels: /cavemenko lite|full|ultra, off via звичайний режим.

Why this matters for the numbers

The 3-arm benchmark in the README was run through agy -p with the ruleset prefixed — i.e. exactly the path documented here. The measured arms are reproducible with the snippet above.

Tests

131 pass, lint pass, version bumped to 2.3.0.

🤖 Generated with Claude Code

---

[devin-ai-integration]

🔍 Devin Review: 2 flags

Not posted on this PR by your GitHub settings — view them in Devin Review. (Configure)

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6d88c13557

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Preserve the skill directory in the manual install

Copying SKILL.md directly into ~/.hermes/skills/ produces ~/.hermes/skills/SKILL.md, but skill discovery expects each skill to be a directory containing that file (as this repository itself does with skills/cavemenko/SKILL.md). Users taking the documented fallback will therefore not have cavemenko discovered; direct them to copy the cavemenko directory or create ~/.hermes/skills/cavemenko/SKILL.md.

Useful? React with 👍 / 👎.