v0.3.0 — OpenClaw adapter

Second host adapter beyond Claude Code. Minor bump, backward-compatible.

Added

• OpenClaw init adapter — python tools/engramory_init.py openclaw --install-skill bootstraps an OpenClaw workspace (default ~/.openclaw/workspace): a marked block in AGENTS.md (loaded every session), the protocol under .agents/skills/engramory (OpenClaw auto-discovers it), and a separate .engramory-memory/ store. The init tool now takes a host arg (codex | openclaw) over a shared, idempotent core; the two adapters use distinct markers and coexist in one AGENTS.md.
• adapters/openclaw/README.md — honest reliability model: the index cap is rules + engramory_check.py. OpenClaw's deterministic deny path is a before_tool_call plugin (TypeScript), not Engramory's Python shell hook (verified against the OpenClaw docs), so the guard does not drop in — a hard cap there means writing that plugin (not shipped here).

Changed (docs)

• PORTING.md host table corrected/extended (Cursor → alwaysApply + auto .agents/skills; added OpenClaw and Trae). The cap "degradation ladder" now lists the real, non-interchangeable per-host deny mechanisms — Hermes pre_tool_call, Cursor preToolUse (new/flaky), OpenClaw before_tool_call plugin, Trae none — each cross-checked against the host's 2026 docs. Only the Claude Code hook is written and tested here.

Tests

• +2 (91 total): OpenClaw bootstrap, and Codex + OpenClaw coexisting with distinct markers.

Full changelog: https://github.com/tinqiao-oss/engramory/blob/master/CHANGELOG.md