← ClaudeAtlas

skill-triagelisted

Pick the right Claude Code skill for a task. Ranks installed skills, emits a routing plan, and falls back to web discovery if nothing matches. Use before non-trivial tasks (3+ steps, architectural decisions, or destructive operations).
karankumar24/skill-triage · ★ 0 · AI & Automation · score 78
Install: claude install-skill karankumar24/skill-triage
# Skill Triage Routing meta-skill. Decides how to approach a task. Prefers skills the user already has installed. When nothing matches, searches a short allow-list of curated registries (and, as a last resort, the open web with URL verification) to suggest skills the user could install. Never recommends an installed skill just because it exists. ## What this skill does Given a task, produce a short, structured recommendation: 1. Restated task (one sentence) 2. Complexity tier (simple / medium / complex / high-risk) 3. Risk flags (destructive, secrets, prod, irreversible, mass edit) 4. Relevant installed skills (ranked, with one-line "why") 5. Skills to avoid (with reason — prevents misfires) 6. Recommended order, grouped by phase: **pre** (plan / brainstorm / research) → **impl** (build / fix) → **post** (review / test / docs / ship) 7. Suggested slash-command invocations 8. Verdict: **proceed directly** | **proceed with skill(s)** | **stop and ask** When the answer is "no skill needed," collapse to a single line — do not pad the template with empty sections. ## When to invoke **Always run before:** - Any task with 3+ steps or cross-file work - Any architectural decision - Anything destructive (deletes, force-push, schema changes, migrations on prod, secret writes) - Any task where multiple plausible skills exist (e.g., a `review` skill vs. a code-quality auditor vs. a security review) **Skip for:** - Trivial one-step edits ("rename this var", "add a console.log") - P