Hunter Bown
06612495fc
Release-preparation checkpoint for v0.8.51 (workspace + npm bumped to 0.8.51). Added: - Arcee AI direct provider: [providers.arcee], ARCEE_API_KEY/BASE_URL/MODEL, CLI auth, provider + model picker, registry. Default direct-API model is trinity-large-thinking (reasoning, 262K ctx/out); preview + mini selectable. Cloudflare-WAF-safe opening turn (benign read-only tool surface, system-prompt payload splitting) and reasoning_content replay on tool-call turns. - Expanded model catalog (qwen3.6 flash/plus/max-preview, Xiaomi MiMo v2.5 chat/ASR/TTS); provider-aware model picker with per-provider saved models. Changed: - Auto-compaction is percentage- and model-aware (compaction_threshold_for_model_at_percent; default 80%; auto-enable for <=256K windows, opt-in for 1M models). - Provider/gateway HTTP errors sanitized (HTML/WAF interstitials collapsed, 401/403 split into authentication vs authorization). Removed: - The session cycle / checkpoint-restart system: /cycles, /cycle, /recall, recall_archive tool, cycle_manager, cycle-handoff prompt, sidebar cycle lines, EngineConfig.cycle / Event::CycleAdvanced / seam cycle thresholds. Fixed: - Orphaned assistant 'blue dot' role glyph on whitespace-only turns. - Sidebar mouse-wheel scroll leaking into the transcript. - Sidebar hover tooltip overlap + warning-orange styling. - README Constitution description corrected to match prompts/base.md. - Repaired release-blocking unit/integration tests after the refactors. Preflight: cargo fmt clean, workspace builds, 3903 tui tests pass (1 known flaky MCP SSE test under parallel load, passes in isolation).
crates/tui/tests/
Integration tests for the TUI binary. Per CONTRIBUTING.md, each crate's
integration tests live in its own tests/ directory; the repository-root
tests/ directory is unused.
Mock LLM client (integration_mock_llm.rs)
crates/tui/src/llm_client/mock.rs provides a MockLlmClient that implements
the LlmClient trait by replaying queue-driven canned responses and capturing
every outgoing MessageRequest. Tests mock at the trait boundary — never
at the reqwest HTTP layer — because the trait is the durable abstraction the
runtime is meant to depend on.
Coverage today exercises the trait surface end-to-end:
- streaming turn loop
- reasoning-content replay across tool-call rounds (V4 §5.1.1, the bug that broke v0.4.9-v0.5.1)
- tool-call round-trip with chunked input JSON
- multi-tool-call ordering inside a single turn
- compaction-style non-streaming
create_message - sub-agent style independent parent/child mocks
- capacity-gate observation of a captured request before stream drain
Four full-engine tests (engine_full_*) are #[ignore]-marked. They unblock
when core::engine::Engine is refactored to take Arc<dyn LlmClient> instead
of a concrete Option<DeepSeekClient>. See the comment block at the bottom of
integration_mock_llm.rs for the exact refactor surface.
--record mode for deepseek eval
The offline deepseek eval harness now accepts --record <DIR>. When set,
each tool step appends one JSON Lines record to <DIR>/<scenario>.jsonl
(default scenario: offline-tool-loop.jsonl). Each line is a self-contained
JSON object with the schema:
{ "request": { "step": "list_dir", "kind": "List" },
"response_events": [ { "type": "ok", "output": "…" } ] }
The mock LLM client (crate::llm_client::mock) replays these fixtures by
mapping each response_events array onto a canned Vec<StreamEvent>. Drop
generated fixtures into crates/tui/tests/fixtures/ so they ride the repo and
feed the mock in CI.
Quick example:
cargo run --bin deepseek -- eval --record crates/tui/tests/fixtures
cat crates/tui/tests/fixtures/offline-tool-loop.jsonl | jq .
The scenario name is sanitized to [A-Za-z0-9_-] before forming the filename,
so unusual scenario strings stay portable across platforms.