メニュー
In-Context Learning — select and inject worked examples (demos) before executing recurring task types. Load when a task matches a known pattern family.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Context engineering — memory navigation, textual graph traversal, progressive disclosure, recovery. Load when assembling context for a task or when lost.
Implementation-side security discipline — headers, dependencies, secrets, authz. Load for hardening work (sec-ops domain).
Commit conventions, branching, anchors in messages, and the git safety rules. Load for any git operation beyond status/diff.
Anti-generic-AI design discipline — bans, tells, and the slop test. Load for ANY new UI surface or visual redesign. Inspired by Impeccable (Apache 2.0).
Motion & animation discipline — purpose, performance, reduced-motion. Load when work involves animation, transitions or scroll effects.
Microcopy discipline — buttons, errors, empty/loading/success states, alt text. Voice and tone come from the project's memory, not from this skill.
| name | icl |
| description | In-Context Learning — select and inject worked examples (demos) before executing recurring task types. Load when a task matches a known pattern family. |
Injecting 1–2 matched demos before a task measurably beats zero-shot on recurring patterns. The pool lives in .berserqir/skills-resources/icl/demos/ (hub, on-demand regime).
| Resolution | Content | Use for |
|---|---|---|
| task-level | full trajectory of a completed similar task (goal → steps → outcome) | complex multi-step tasks |
| step-level | snippet of one recurring step (e.g., "update memory-medium after task") | protocol steps agents get wrong |
core/protocols/context-budget.md.---
tags: [area, pattern-family]
outcome: success | anti-example
source: manual | promoted-instinct | eval-failure
---
# Goal
# Trajectory (condensed ReAct: thought → action → observation)
# Outcome & verification
Eval failures and recurring errors become outcome: anti-example demos — injected alongside a success demo when the task family has a known failure mode ("this looks right but fails because…").
Demos enter the pool by: manual curation · instinct promotion (confidence ≥ threshold + eval pass — see plan F4) · eval failure capture. Every demo carries source.