core/cli: show estimated context tokens and usage source tags

/status adds context_tokens~=(est); docs note api/estimate/mixed labels.

CLAUDE.md

@@ -58,7 +58,7 @@ Async SQLAlchemy + aiosqlite. `MemoryStore`: schema init (+ lightweight SQLite `
 - **`ChatAgent`**: optional `MemoryStore` (persist on success only); `stream`; system prompt; `pending_retry_text` + `retry()`.
 - **`/compact`**: soft-compact tool dumps → model summary (no tools) → **new** session seeded with summary message.
 - Events: text_delta, assistant_message, tool_call/result, max_rounds, **error** (`retryable`/`source`), **cancelled** (`reason`), **usage** (`TokenUsage`).
-- Usage: API `usage` from completions (stream with `include_usage`); `ChatAgent.session_usage` / `last_turn_usage`; CLI end-of-turn + `/status`.
+- Usage: API `usage` from completions (stream with `include_usage`); **char≈token fallback** (~4 chars/token) when omitted; `ChatAgent.session_usage` / `last_turn_usage`; CLI end-of-turn + `/status` (marks `(est)` / `(api+est)`).
 - Config ``[agent]``: `system_prompt`, `max_tool_result_chars`, `parallel_tools`, `confirm_destructive`, `path_denylist`, `max_context_chars`.
 
 ### Tools (`tools/`)
@@ -101,7 +101,7 @@ Basedpyright `recommended`. Ruff includes `ANN` (private return types `ANN202` i
 - **Phase D (context quality)**: soft char budget, request compact, `/compact`, richer errors/cancel, workspace sessions. Context size is **char estimate** plus optional API usage when reported.
 - **No local tokenizer stage** for now.
 - **Phase E**: tooling depth (grep/glob, VCS backends; prefer `edit_replace` / `edit_lineno` over model-generated patches).
-- **Phase F (providers + usage v2)**: capture response/stream `usage` — **in progress / landing**; session/turn totals; `/status` + end-of-turn line. Optional later: cost, tokenizer-backed estimates.
+- **Phase F (providers + usage v2)**: API `usage` + char-based estimate fallback; session/turn totals; `/status` + end-of-turn. Optional later: cost, real tokenizer.
 - **Phase G–H**: CLI polish, hardening; then multi-tenant platform (`router/`, auth, sandboxed tools).
 
 ## Commit messages

src/plyngent/cli/repl.py

@@ -60,8 +60,11 @@ def _cmd_status(state: ReplState) -> None:
     pending_disp = "yes" if pending else "no"
     ctx_chars = estimate_messages_chars(state.agent.messages)
     ctx_budget = state.agent.max_context_chars
+    from plyngent.agent.usage import chars_to_tokens
+
     session_u = state.agent.session_usage
     last_u = state.agent.last_turn_usage
+    ctx_est_tokens = chars_to_tokens(ctx_chars)
     click.echo(
         f"provider={state.provider_name}  model={state.model}\n"
         f"session={state.session_id}  messages={len(state.agent.messages)}  "
@@ -69,6 +72,7 @@ def _cmd_status(state: ReplState) -> None:
         f"tools={'on' if state.tools_enabled else 'off'}  "
         f"rounds={state.max_rounds}  stream={'on' if state.agent.stream else 'off'}\n"
         f"context_chars={ctx_chars}/{ctx_budget}  "
+        f"context_tokens~={ctx_est_tokens} (est)  "
         f"tool_result_max={state.agent.max_tool_result_chars}\n"
         f"usage_session={session_u.format_line()}\n"
         f"usage_last_turn={last_u.format_line()}\n"