v1.0.0

v1.0.0 — behavior-enhancer-generic

Behaviour discipline for coding agents, shipped as a cross-host dual-form product: FULL plugin/hook adapters where the host exposes real interception, and a host-agnostic SOFT skill where it does not.

Install on Claude Code

/plugin marketplace add Zoria-Lind/behavior-enhancer-generic
/plugin install behavior-enhancer@zoria-behavior

Other hosts: see the per-host README under adapters/ (Codex: config.toml hooks or plugin manifest; Pi: extension; DSH: install the original dsh-behavior-enhancer plugin instead). The host-agnostic SOFT skill is skill/SKILL.md; a packaged copy is also listed free on Agensi.

Adapters in this release

Adapter	Form	Tests
Claude Code	plugin (hooks)	14 / 14 engine-level
Codex	config.toml hooks or plugin manifest	13 / 13 protocol
Pi	pi extension	11 / 11 protocol
DSH	pointer to the original dsh-behavior-enhancer plugin — install that one	—
any other host	SOFT skill + ADAPTIVE self-built adapter guidance	—

Each feature activates independently, based on the host's capability surface — never on user or agent preference. Shipped features include read-before-write veto (read-coverage tracking, /force-read markers), failure convergence, write verification with auto-rollback, a high-risk command gate, concurrency-window veto (F3b), and a status command.

Honest edges

• SOFT is not FULL. The skill cannot intercept anything — it makes the agent want to comply. Hard enforcement ships only as adapters.
• NATIVE is claimed only for hosts where the behaviour was measured; unverified hosts are treated as not having it.
• The design documents (core/, generic-skill_plan_v1.md) are currently Chinese-only; the READMEs are bilingual with English as the primary.
• Install one form per host, never two: a marketplace copy plus a local --plugin-dir / CLAUDE_CODE_PLUGIN_DIRS copy of the same adapter will double-run the same hooks.
• DSH users: install either this repo's adapters/dsh or the original dsh-behavior-enhancer — not both.

MIT licensed.