continuous-improvement-triagelisted

Operationalises spec/project/continuous-improvement/ by triaging portfolio audit findings, classifying them against the available specialist catalog, and dispatching hands-on remediation to the most specialised available Claude agent or skill. Invoke when the user asks to "triage continuous-improvement findings", "classify a portfolio-improvement opportunity", "dispatch findings to specialised agents", "run the quarterly specialist-coverage review", "check whether a finding class needs a new specialist", or equivalent German-language requests. Drives the three-operation loop: audit (periodic specialist-coverage review), update (record decisions and dispatch specialists), close (terminate the triage cycle). Applies the three-recurrence gap-closure rule and records all dispatch decisions in fix-PR Risk / rollout notes. Do not use to perform the hands-on remediation itself—that belongs to the dispatched specialist. Supports resume on re-invocation per `spec/claude/resumable-work/`.
nolte/claude-shared · ★ 0 · AI & Automation · score 76

Install: claude install-skill nolte/claude-shared

# Continuous Improvement Triage Implements `spec/project/continuous-improvement/` §Specialist dispatch, §Portfolio gap closure, and §Continuous loop and quarterly coverage review. The skill is the generalist orchestrator that classifies findings, dispatches the hands-on work to the most specialised available Claude agent or skill, and verifies that every fix PR carries the required audit trail. It never performs the specialist remediation itself when a matching specialist exists. ## Why this is a skill, not an agent - **User confirmation gates every dispatch decision.** Finding classification, specialist-match confirmation, and gap-closure decisions all require mid-flow dialogue; an agent's fire-and-forget shape misses them. - **Orchestrator pattern (per `skill-vs-agent`).** The work is *classify, dispatch, verify*—not edit. The dispatched specialist does the mutating work; this skill stays in the main thread and chains other skills (`pull-request-create`, optionally `pull-request-merge`). - **Quarterly cadence requires standing-session instructions.** The specialist-coverage review surfaces decisions that accumulate across multiple prompts; a skill's persistent instruction context fits this naturally. - Counter-dimension considered: a narrow agent could handle the classification step in isolation and gain context-window protection, but every downstream lane (dispatch, gap-closure decision, PR annotation) is interactive—keeping the whole flow in one skill is simpler than s