docs: sync README files with latest repo state

plugin/src/README.md

@@ -1,5 +1,5 @@
 # claude_smart (plugin/src)
-Description: Python package powering the claude-smart plugin — hook handlers that buffer Claude Code / Codex session activity, publish it to a local Reflexio backend, and inject the learned skills/preferences back into future sessions.
+Description: Python package powering the claude-smart plugin — hook handlers and CLI flows that buffer Claude Code / Codex / OpenCode session activity, publish it to a local Reflexio backend, and inject learned skills/preferences back into future sessions.
 
 > User-facing install/usage docs live in [`../README.md`](../README.md). This is the code map for the `claude_smart` package.
 
@@ -8,7 +8,7 @@ Description: Python package powering the claude-smart plugin — hook handlers t
 | File | Purpose |
 |------|---------|
 | `hook.py` | Hook dispatcher. Parses the stdin JSON payload and routes each event (Setup, SessionStart, UserPromptSubmit, PreToolUse, PostToolUse, Stop, SessionEnd) to its handler in `events/`. |
-| `cli.py` | `claude-smart` CLI: `install`, `uninstall`, `show`, `learn`, `restart`, `dashboard`, `clear-all`. |
+| `cli.py` | `claude-smart` CLI: `install`, `uninstall`, `show`, `learn`, `restart`, `dashboard`, `clear-all`; installs host-specific Claude Code, Codex, or OpenCode integration files. |
 | `state.py` | Per-session JSONL buffer at `~/.claude-smart/sessions/{session_id}.jsonl`; high-water mark for idempotent publish retries. |
 | `publish.py` | Drains the session buffer, calls the Reflexio adapter, and stamps the watermark. Used by Stop, SessionEnd, and `learn`. |
 | `reflexio_adapter.py` | Thin wrapper over `reflexio.ReflexioClient` — isolates import failures, 5s HTTP timeout, injects/queries learnings, degrades gracefully when the backend is down. |
@@ -29,29 +29,30 @@ Description: Python package powering the claude-smart plugin — hook handlers t
 | File | Purpose |
 |------|---------|
 | `env_config.py` | Parses `~/.claude-smart/.env` (`REFLEXIO_URL`, `REFLEXIO_API_KEY`, `CLAUDE_SMART_READ_ONLY`, local embedding/CLI flags). |
-| `runtime.py` | Host detection (Claude Code vs Codex); shared agent version. |
+| `runtime.py` | Host detection (Claude Code, Codex, OpenCode); shared agent version. |
 | `ids.py` | Session / project ID generation and resolution. |
-| `context_inject.py`, `context_format.py`, `query_compose.py`, `cs_cite.py` | Build search queries, format learned skills as markdown, inject into context, format citations. |
+| `context_inject.py`, `context_format.py`, `query_compose.py`, `cs_cite.py` | Build search queries, format learned skills as markdown, inject into host context, format citations. |
 | `stall_banner.py` | User-facing message when the Reflexio provider hits an auth/billing stall. |
 | `optimizer_assistant.py` | Claude-code CLI agent that extracts Reflexio optimization hints. |
 | `hook_log.py`, `internal_call.py` | Structured JSON logging to `~/.claude-smart/hook.log`; detect internal/test invocations to skip learning. |
 
 ## Architecture
 
 ```
-SessionStart -> bootstrap Reflexio backend (8071) + Next.js dashboard (3001)
-PostToolUse  -> state.py buffers each tool call to ~/.claude-smart/sessions/{id}.jsonl
-Stop/SessionEnd -> publish.py -> reflexio_adapter -> Reflexio extracts playbooks + preferences
-UserPromptSubmit / PreToolUse -> context_inject pulls relevant learnings back into context
+Claude Code/Codex hooks -> hook.py -> events/* -> state.py buffer
+OpenCode plugin -> ../opencode/* -> claude_smart CLI/publish paths
+Stop/SessionEnd/learn -> publish.py -> reflexio_adapter -> Reflexio extracts playbooks + preferences
+UserPromptSubmit / PreToolUse / OpenCode request path -> context_inject pulls relevant learnings back into context
 ```
 
-- **Dual-host** — the same package runs under Claude Code (native slash commands) and Codex (shell-script fallbacks in `../scripts/`); host shape is normalized in `runtime.py`.
+- **Multi-host** — the same package runs under Claude Code (native slash commands), Codex (shell-script fallbacks in `../scripts/`), and OpenCode (npm plugin bridge in `../opencode/`); host shape is normalized in `runtime.py`.
 - **Offline-resilient** — if Reflexio is unreachable, interactions stay buffered and drain on the next successful publish.
-- **Hooks registered** in `../hooks/hooks.json` (Claude Code) and `../hooks/codex-hooks.json` (Codex), dispatched through `../scripts/hook_entry.sh`.
+- **Hooks/bridges registered** in `../hooks/hooks.json` (Claude Code), `../hooks/codex-hooks.json` (Codex), and OpenCode config (`opencode.json` / `~/.config/opencode/opencode.json`) pointing at `../opencode/dist/server.mjs`; Claude/Codex events dispatch through `../scripts/hook_entry.sh`.
 
 ## Requirements / Problems to Avoid
 
 - **All learned state is external to the repo** — sessions buffer under `~/.claude-smart/`, Reflexio state under `~/.reflexio/`. Don't write generated runtime artifacts inside the plugin root.
 - **Backend lives at `http://localhost:8071`**, dashboard at `http://localhost:3001`; both are long-lived across sessions (no hard shutdown on SessionEnd by default).
-- **`CLAUDE_SMART_READ_ONLY`** skips publishing entirely — respect it in any new publish path.
+- **`CLAUDE_SMART_READ_ONLY`** skips publishing entirely — respect it in any new publish path, including host bridges.
+- **OpenCode extraction runs isolated** — `CLAUDE_SMART_OPENCODE_MODEL` can override the model used by internal `opencode run --pure` extraction; do not assume a project config is loaded for that subprocess.
 - **Keep handlers fast** — tool-use hooks run on a 10–15s timeout; heavy work belongs in the backend, not the hook.