fe-claude-hp

Cross-repo atlas for HeroPlus Remote Key Injection (RKI / BDK / IPEK / KCV / DUKPT / APC / kif-bridge / Sunmi portal / KPay-KDP / MYHSM). Auto-fires on key-injection / Sunmi key_type / RKIS protocol / KSN / kif-bridge / MYHSM queries.

Research-backed rules for creating, auditing, modifying, or designing Claude Code skills, CLAUDE.md files, rules files, hooks, subagents, or any system-prompt instruction file. Auto-fires on skill / CLAUDE.md / hook / subagent / instruction-file authoring intents.

Intent-first commit message style — each commit-body point names the problem the change solves, not what the diff shows. Auto-fires when generating any commit message or running git commit; also provides /commit-message:staged, which commits your staged changes with a message drafted on a read-only Sonnet subagent.

Universal FE code-craft skills presenting under each repo's /architecture:* namespace beside its own project architecture skill — comment hygiene (/architecture:comment-style) and single-responsibility / concern-separation (/architecture:code-structure). Auto-fires on writing, editing, or reviewing source and nudges on size-band growth (Kotlin/TypeScript/JS).

The FE self-learning harness engine — dimension context router, shadow capture gate, payment security floor (secret-scan, payment firewall, self-contained commit-boundary git hooks). Enabled per repo via enabledPlugins; per-repo committed .claude/self-learning.json tunes captureMode + additive guard/lexicon extensions; routes, dimensions, and telemetry stay repo data. Ships /self-learning:doctor, /self-learning:routes, /self-learning:shadow-rate.

Write-back coherency — persist a session's load-bearing discoveries into their single canonical home (repo docs / per-project memory / a shared skill), via a clean-context curator that routes facts while the human gates the commit. Auto-fires at a work boundary or on capture / persist / 'write down what we learned or corrected' intents.