eval-audit
**Locale**: All templates in this spec are written in English. Detect the user's language from the session and translate user-facing text at display time per…
Triggered by the SessionStart hook (`hooks/scripts/session-tracker.js`), which compares the current SkillCompass version against the last recorded version. On first install, reinstall, or version change the hook injects a context message instructing Claude to load this file and
$ npx -y skills add Evol-ai/SkillCompass --agent claude-codeHow it fires
How this command gets triggered: by you, by Claude, or both.
/post-install-onboardingContext preview
What this command does when you run it.
Triggered by the SessionStart hook (`hooks/scripts/session-tracker.js`), which compares the current SkillCompass version against the last recorded version. On first install, reinstall, or version change the hook injects a context message instructing Claude to load this file and
Triggered by the SessionStart hook (`hooks/scripts/session-tracker.js`), which compares the current SkillCompass version against the last recorded version. On first install, reinstall, or version change the hook injects a context message instructing Claude to load this file and follow it on the user's first interaction.
Do not wait for a slash command. Execute the steps below in order as soon as the session starts.
🧭 SkillCompass Evaluate skill quality, find the weakest area, and improve it. Also tracks usage to help you spot idle or risky skills.
Silently run skill discovery — scan immediate children `*/SKILL.md` in each skill root (same as `/setup` Step 3; do NOT recurse) — and run a D1+D2+D3 quick scan on every discovered skill. Save the results to `.skill-compass/setup-state.json`. Then show one of the two result summaries below.
If any risks found:
Scanning installed skills...
Found {N} skill(s){, including M collections if any}.
{K} have security or structural risks; the rest passed the quick scan ✓
[View risky skills / Continue]If all clean:
Scanning installed skills...
Found {N} skill(s){, including M collections if any}, all passed the quick scan ✓
[Continue]Check whether `~/.claude/settings.json` already has a `statusLine` configured.
If NO existing statusLine:
SkillCompass tracks skill usage automatically. When suggestions exist, the status line shows 🧭 N pending — type /skillcompass to view. [Enable status line 🧭 / Skip]
If the user chooses Enable, offer two modes:
[Minimal — just the 🧭 hint / Full HUD — model, context, and more]
If YES existing statusLine: skip silently — do not overwrite the user's configuration.
✓ Setup complete. SkillCompass runs in the background: · Tracks skill usage frequency · Surfaces idle or problematic skills · Shows 🧭 in the status line when there are suggestions Type /skillcompass anytime to view and manage skills.
After displaying the finish message, record the current version so the onboarding won't trigger again next session. Use the **Bash** tool to run:
node "${CLAUDE_PLUGIN_ROOT:-.}/hooks/scripts/write-last-version.js"The script is silent on success and on failure (the onboarding must complete even if the sidecar cannot be written).
After onboarding, do NOT show the inbox view. The user was not asking for inbox — they were just starting a session. Return control to whatever the user intended to do.
Evaluate agent skill quality. Find the weakest link. Fix it. Prove it worked.
Repo: Evol-ai/SkillCompass
**Locale**: All templates in this spec are written in English. Detect the user's language from the session and translate user-facing text at display time per…
**Locale**: All templates in this spec are written in English. Detect the user's language from the session and translate user-facing text at display time per…
**Locale**: All templates in this spec are written in English. Detect the user's language from the session and translate user-facing text at display time per…
- **Recommended model: Claude Opus 4.6** (`claude-opus-4-6`). Directed improvement requires understanding complex rubric feedback and generating precise,…
**Locale**: All templates in this spec are written in English. Detect the user's language from the session and translate user-facing text at display time per…
**Locale**: All templates in this spec are written in English. Detect the user's language from the session and translate user-facing text at display time per…