feedback-integratorlisted

Aggregate repeated findings, acceptance failures, eval results, user corrections, and workflow incidents across skill runs to identify recurring failure patterns and propose evidence-backed improvements to skill instructions, routing, references, kernels, evals, or host integration. Use when the user wants the skill system to learn from repeated mistakes, run a retrospective, improve reliability over time, or convert observed failures into regression tests. Do not use to silently self-modify skills, learn from a single weak signal, replace product/research analytics, or apply repository changes without explicit authorization.
CometWeb-io/agent-skills · ★ 1 · AI & Automation · score 74

Install: claude install-skill CometWeb-io/agent-skills

# Feedback Integrator Operate the improvement loop `observed failure -> pattern -> causal hypothesis -> change proposal -> regression test -> re-evaluation`. Do not call ordinary accumulation "learning". ## 1. Normalize observations Accept reviewer/roaster findings, acceptance reports, eval failures, routing misses, user corrections, tool failures, runtime drift alerts, rollout/rollback incidents, and incident notes. Preserve skill version, host/model when known, prompt/task class, timestamp, severity, and evidence. ## 2. Separate incidents from patterns Default threshold: a pattern needs recurrence across at least two independent runs or one clearly severe systemic failure. Do not generalize from one preference correction or one host-specific glitch. Track correlated runs so the same copied failure is not counted as independent evidence. ## 3. Classify root layer Map each pattern to the smallest layer likely to fix it: - `ROUTING` — skill failed to trigger or triggered on near-miss; - `INSTRUCTION` — workflow ambiguity or missing guard; - `REFERENCE` — needed detail not reachable/clear; - `KERNEL` — deterministic invariant not enforced; - `EVAL` — failure mode not pinned by tests; - `HOST` — integration/capability mismatch; - `PROCESS` — orchestration/handoff/state problem; - `RUNTIME` — host/model/version drift, rollout failure, latency/cost regression, or runtime-only incompatibility. Avoid patching SKILL.md when the real defect is an unenforced invariant or host