master

hypothesiserfeat: enhance external command handling in TUI and add tests

2c25d0e5创建于 10 天前历史提交

文件	最后提交记录	最后更新时间
agent_loop.rs	feat: enhance external command handling in TUI and add tests - Refactored external command processing to improve clarity and functionality. - Introduced new functions for expanding commands and handling user arguments. - Added unit tests to ensure correct behavior of external command expansion. - Updated Cargo.lock to include new dependencies: libc and tracing.	10 天前
error.rs	feat: initial commit with core features	2 个月前
input.rs	feat: implement pending user message handling in chat state and gateway - Added support for managing pending user messages within the chat state and gateway runtime. - Introduced methods for enqueueing and discarding pending user messages, enhancing user interaction during ongoing tasks. - Updated the agent loop to drain pending user messages and integrate them into the chat flow. - Enhanced session management to accommodate pending messages, ensuring a smoother user experience. - Updated relevant tests to verify the functionality of pending user message handling.	1 个月前
kvcache.rs	cargo fmt	1 个月前
lib.rs	optimize: add proactive token estimation and config validation with URL optimization Co-Authored-By: GLM-5	17 天前
loop_state.rs	fix(core): concurrent batch correctness, finish nudge gating, prune projection Four fixes to the agent loop: - Mid-batch stop no longer discards already-executed results. A stop rule matching partway through a batch previously broke out of the record loop, stranding the remaining (already-run) tool_use blocks without paired tool_results and dropping their side effects from history. Now every result is recorded first and the stop is applied only after the whole batch is folded in. - Completion checklist gates on a tool having actually executed in the run (new LoopState.tool_executed), not merely on tools being visible — a tool-less conversational turn is no longer charged an extra verification round-trip. - Batch concurrency is opt-in via the declared EffectProfile: a batch runs concurrently only when every call is a pure reader (reads fs/network, no writes, no side effects). Writers, bash, todo_write, join_subagent, interactive prompts, and unknown tools fall back to the sequential lane, replacing the narrow same-file_path guard that let them race. - Stale tool-output pruning runs on the prompt projection (a clone) instead of mutating canonical history, so lookback/post-mortem keep the full record. Tests: 4 added (mid-batch stop, conversational no-nudge, is_parallel_safe allow/deny). 29 pass. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>	12 天前
outcome.rs	feat: initial commit with core features	2 个月前
runtime.rs	optimize: add proactive token estimation and config validation with URL optimization Co-Authored-By: GLM-5	17 天前
runtime_support.rs	feat(tool/bash): add interactive command detection to prevent execution requiring user interaction Co-Authored-By: GLM-5	15 天前
snapshot.rs	feat: initial commit with core features	2 个月前
suspend.rs	feat(core): concurrent tool batch, verified finish, plan on control path Loop-level execution quality, independent of the inner model: - Independent tool calls in a turn run concurrently while preserving call-order history; a suspending join_subagent is ordered last and same-path side-effecting calls fall back to sequential, so a batch can neither strand a committed mutation across a suspend nor race two writers on one file. Several suspends in one turn each resolve on resume (no stranded tool_use blocks). - Verified finish: the first stop is not accepted; a one-shot completion checklist re-anchors on the original task and only a second stop completes. Gated to runs that expose tools (a conversational turn is not nudged). - Plan on the control path: todo_write exposes a reader; open items are re-injected each turn and consulted before a stop is accepted. - Error classification: transient (stream/io/timeout/throttle) vs permanent; a repeated identical failing OR successful call injects a strategy-change nudge; malformed tool names are normalized before rejection. - On the final turn tools are withheld so the model commits a final answer rather than a cut-off tool call. Stale tool-result bodies beyond a recent window are pruned to a marker (token-keyed). - Declares futures-util in crates/core for the concurrent join. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>	13 天前
token_estimator.rs	fix(tui): input auto-wrap, Alt+Enter newline, and intelligent arrow key navigation By the way, apply cargo fmt for code style consistency. Co-Authored-By: GLM-5	14 天前