Conversation
Documentation build overview
15 files changed ·
|
Documentation build overview
15 files changed ·
|
| assert any(isinstance(event, UsageEvent) for event in sink.events) | ||
|
|
||
|
|
||
| def _model_response_events(index: int, *, text: bool = False, finish: bool = False): |
There was a problem hiding this comment.
Could we simplify the tests here? The core behavior is fairly small, but the new coverage expands into several overlapping parameterized cases across CLI and Dashboard. I think a few focused cases for response counting, budget interruption, and preserving finish, plus one dashboard smoke test, should be sufficient.
There was a problem hiding this comment.
Reduced the ten test instances to four in 854d1b8: response counting/deduplication, CLI interruption, preserving finish, and one Dashboard smoke test. The reasoning/tool-only, duplicate-usage and cleanup checks remain. Full suite: 587 passed, 3 optional-dependency skips; pre-commit and both documentation builds also pass.
| if not text: | ||
| return "" | ||
| self.final_response = text | ||
| self.turns += 1 |
There was a problem hiding this comment.
Is this line an error or an unused assignment?
There was a problem hiding this comment.
The removal is intentional. agentMessage rendering now only handles text; _set_usage counts completed responses, including reasoning/tool-only responses. Restoring this increment would double-count responses that also emit text. The four existing budget regression tests pass.
| ``100``。各后端对一轮的定义和上限的执行方式不同，报告的 ``turns_used`` | ||
| 与执行预算时使用的计数不一定相同： | ||
|
|
||
| .. list-table:: 各后端的轮数统计与限制 |
There was a problem hiding this comment.
It's still hard to understand. please use understandable description.
There was a problem hiding this comment.
Replaced the table with a short explanation and a two-file-read example in 0abaca2. Both guides now explain what counts as a turn and why Claude's reported count can differ from its limit. Runtime behavior is unchanged; the existing tests and both documentation builds pass.
dedbb9e to
0abaca2
Compare
5 participants
Codex previously counted nonempty text items, so reasoning/tool-only responses could bypass
--max-turns. The planner now counts distinct usage updates, uses that count in CLI and Dashboard, and preserves a completedfinishwhen stopping at the budget.The English and Chinese guides explain turns with a two-file-read example. They distinguish RPent's response count from Claude's tool-use limit without changing Claude's runtime behavior.
Rebased onto current main
eecf2069, preserving the complete original changes.Validation