skill-activate

Star the repo: https://github.com/realsahilsaini/skill-activate

Claude Code can miss relevant skills when responding to prompts.

skill-activate installs a forced-eval hook and rewrites skill descriptions so Claude evaluates and activates relevant skills before every response.

One-line install

npx skill-activate install

Revert to original state anytime:

npx skill-activate uninstall

Follow development and contribute: https://github.com/realsahilsaini/skill-activate

Reliability impact

• Before: skill activation can be inconsistent for relevant prompts.
• After install: prompts are forced through explicit skill evaluation and activation before implementation.

How it works

• Rewrites each skill description into a directive format with clear triggers and a hard activation instruction.
• Installs a UserPromptSubmit hook that injects a mandatory evaluation-and-activation sequence before every prompt.
• Patches Claude settings safely and idempotently, then validates everything with a readable report.

Commands

| Command | What it does | | --- | --- | | npx skill-activate install | Full setup: rewrite descriptions, install hook, patch settings, validate | | npx skill-activate scan | Detect and list installed SKILL.md files only | | npx skill-activate uninstall | Remove hook wiring and restore all .bak files under skills | | npx skill-activate status | Verify if hook is present and correctly wired |

How this works under the hood

skill-activate uses Claude Code's UserPromptSubmit lifecycle hook to prepend a mandatory instruction block on each user message. That block forces Claude to:

1. Evaluate each available skill (YES or NO with reason)
2. Activate every YES skill through Skill() immediately
3. Only then continue with implementation

Reference: Anthropic Claude Code hooks docs: https://docs.anthropic.com/en/docs/claude-code/hooks

Output example

skill-activate setup report
───────────────────────────────────────
✓ Claude Code detected at ~/.claude
✓ Hook installed: ~/.claude/hooks/skill-forced-eval-hook.sh (executable)
✓ settings.json patched (UserPromptSubmit wired)
✓ 3 skills found and descriptions rewritten:
    → express-skill.md
    → mongoose-skill.md
    → jwt-auth-skill.md
✓ Backups saved as .bak files

Activation consistency improved with forced-eval hook.
───────────────────────────────────────
Restart Claude Code to apply changes.

Development

npm install
npm test
node bin/cli.js install

How to contribute

1. Fork the repo and create a feature branch.
2. Run tests locally (npm test) before opening a PR.
3. Add or update tests for behavior changes.
4. Keep changes local-only (no external API dependencies).
5. Open a PR with a clear before/after summary.

If this helped you, please star and watch the repo for updates:

• Star: https://github.com/realsahilsaini/skill-activate/stargazers
• Watch: https://github.com/realsahilsaini/skill-activate/watchers
• Issues: https://github.com/realsahilsaini/skill-activate/issues

Compatibility

• Node.js 18+
• macOS and Linux (native shell hook execution)
• Windows with WSL for shell hook execution

Troubleshooting

• Error: Unknown skill: ... : This usually means the skill id was normalized (for example, algorithmic-art changed to algorithmic art). Re-run npx skill-activate@latest install to apply the newest compatibility behavior. Version 0.1.2+ automatically creates compatibility aliases for hyphenated and underscored skill ids.

License

MIT