feat(cli): add stats command

[yan-ad]

Summary

• add crabcode stats with an OpenCode-compatible terminal layout
• support --days, --tools, --models [N], and --project [PROJECT]
• aggregate persisted sessions, messages, output tokens, cost, model usage, and tool calls
• add CLI parsing, aggregation, formatting, and rendering tests

Validation

• cargo fmt
• focused unit and CLI parser tests
• cargo check
• manual crabcode stats output comparison against opencode stats

Notes

Crabcode currently does not persist input/cache token breakdowns or per-message model cost, so those fields remain zero until that data is recorded. Existing output token, session cost, model, and tool-call data are reported from SQLite.

[Blankeos]

super nice! thanks!

[Blankeos]

Merge confidence: 3/5 (feat/stats-command -> main)

Stats CLI looks good. Blocking issues are on the live Anthropic stream path, plus this branch now needs a rebase.

Rebase first

GitHub reports CONFLICTING vs current main (at least src/aisdk/README.md, compatible.rs, openai.rs, response.rs after #39/#41/#42). Please rebase onto main before anything else.

Blockers

1. Anthropic message_delta drops stop_reason. Real Anthropic events always include usage. The new branch returns ChunkType::Usage and never calls anthropic_message_delta, so max_tokens / refusal / finish-reason End are lost. Existing tests omit usage, so they still pass.

   Probe with usage + stop_reason=max_tokens got Usage(...) instead of Incomplete. In aisdk/response.rs, Incomplete becomes Failed and aborts the step — so truncated/refusal turns now look like a clean message_stop success.

2. Anthropic usage double-count. message_start emits partial usage; message_delta emits final. Chat::record_usage saturating-adds, so input/cache/cost can be ~2x the bill.

Non-blocking

• No DB migration (usage lives in existing messages.parts JSON).
• Compatible SSE can emit Usage + End together (good).
• Unconditional stream_options.include_usage: true is fine for Ollama/Groq today, but some strict proxies still 400 — consider gating.
• ACP persists usage with cost hardcoded 0.0 and pushes extra parts instead of merging.
• cargo fmt --check, aisdk boundary, cargo check --tests, and focused stats tests passed.

Checklist

• Rebase onto main
• Emit Anthropic usage without swallowing stop_reason (or attach usage only to the final event)
• Tests that include the usage object: max_tokens / end_turn, and no message_start double-count
• Optional: gate include_usage; ACP cost parity

Fix the Anthropic stop_reason + double-count with those tests → 4/5. Rebase + that → merge-ready from my side.

Thanks for the OpenCode-style layout — the CLI aggregation itself is in good shape.

[yan-ad]

Resolved the requested blockers in b1bce54 and rebased the branch onto current main.

• Anthropic message_delta now emits final Usage followed by the original terminal chunk, preserving max_tokens, refusal, and finish reasons.
• message_start usage is log-only, so partial + final usage is no longer double-counted.
• Added realistic usage-bearing tests for max_tokens, end_turn, and no message_start emission.

Validation: cargo fmt --check, cargo check --tests, focused Anthropic/OpenAI/compatible/stats tests, and AISDK boundary equivalent all pass. Please re-review when convenient.

[Blankeos]

PR Review — #40 feat(cli): add stats command (follow-up)

Branch: feat/stats-command → main
PR: #40
HEAD: b1bce54 fix(aisdk): preserve Anthropic stop reasons
Prior score: 3/5

Merge confidence: 4/5

Anthropic blockers from the last review are fixed with tests. Remaining 1 point is a small rebase vs latest main plus non-blocking ACP/include_usage nits.

---

GitHub comment body

Thanks — b1bce54 addresses both Anthropic blockers.

message_delta now emits Usage then Incomplete/End, and message_start only logs partial usage (not persisted). Tests include the live usage object. Focused tests pass (max_tokens_delta_emits_final_usage_then_incomplete, end_turn_delta_emits_final_usage_then_terminal_reason, message_start_usage_is_not_emitted_or_double_counted).

Please rebase onto current main again. Latest main (4b41a3e #37 completions) conflicts in src/main.rs root_help_lists_completion (help string for completion vs new stats line). GitHub mergeable was MERGEABLE right after your rebase, then main moved.

Non-blocking leftovers: ACP still hardcodes cost 0.0 and pushes extra usage parts; stream_options.include_usage is still unconditional.

No DB migration. Stats CLI still looks good.

Score: 4/5 (was 3/5). Rebase the help test → merge-ready from my side.

---

What changed since last review

Item	Last review	Now
Rebase	CONFLICTING vs main	Rebased; then main moved again (#37) — small conflict in root_help_lists_completion
Anthropic stop_reason swallowed by Usage	blocker	Fixed — anthropic_stream_chunks can emit both
message_start double-count	blocker	Fixed — start is log-only
Tests with usage payload	missing	Added
include_usage always on	nit	unchanged
ACP cost 0.0 / extra parts	nit	unchanged
DB migration	none	none

Checks run

Check	Result
cargo fmt --all -- --check	pass
scripts/check-aisdk-boundary.sh	aisdk boundary ok
Anthropic usage tests + stats/CLI/record_usage	10 passed

Confidence bumps

If you do	New score
Rebase onto main and fix root_help_lists_completion	4.5/5
Above + ACP merge/pricing parity, or gate include_usage	5/5

[Blankeos]

LGTM thanks @yan-ad !