Bump version to 1.6.3
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Opus 4.6
parent
4ba9706e35
commit
59cfd0e2f5
@@ -13,15 +13,8 @@ use crate::{
|
||||
},
|
||||
};
|
||||
|
||||
/// Minimum number of tokens users' input should have before kicking off input detection
|
||||
/// to switch from AI input to command input.
|
||||
/// This could be tuned.
|
||||
const MINIMUM_COMMAND_DETECTION_TOKEN_LENGTH: u8 = 2;
|
||||
|
||||
/// Minimum number of tokens users' input should have before kicking off input detection
|
||||
/// to switch from command input to AI input.
|
||||
/// This could be tuned.
|
||||
const MINIMUM_NATURAL_LANGUAGE_DETECTION_TOKEN_LENGTH: u8 = 2;
|
||||
/// Minimum number of characters the input buffer must have before classification runs.
|
||||
const MINIMUM_CLASSIFICATION_CHAR_LENGTH: usize = 3;
|
||||
|
||||
/// The percentage of input tokens that can be recognized as a natural language word before
|
||||
/// we consider the input as natural language. This could be tuned.
|
||||
@@ -43,20 +36,32 @@ impl InputClassifier for HeuristicClassifier {
|
||||
let word_tokens = parse_query_into_tokens(input.buffer_text.as_str());
|
||||
let total_word_token_count = word_tokens.len();
|
||||
|
||||
log::info!(
|
||||
"[input-classifier] detect_input_type called: buffer={:?}, word_tokens={}, current={:?}",
|
||||
&input.buffer_text,
|
||||
total_word_token_count,
|
||||
context.current_input_type
|
||||
);
|
||||
|
||||
if total_word_token_count == 1
|
||||
&& is_one_off_natural_language_word_or_prefix(&word_tokens[0].to_lowercase())
|
||||
{
|
||||
log::info!("[input-classifier] → one-off NL word, returning AI");
|
||||
return InputType::AI;
|
||||
}
|
||||
|
||||
if is_likely_shell_command(&input, total_word_token_count).await {
|
||||
log::info!("[input-classifier] → is_likely_shell_command=true, returning Shell");
|
||||
return InputType::Shell;
|
||||
}
|
||||
|
||||
self.classify_input(input, context)
|
||||
let result = self.classify_input(input, context)
|
||||
.await
|
||||
.map(|result| result.to_input_type())
|
||||
.unwrap_or(context.current_input_type)
|
||||
.unwrap_or(context.current_input_type);
|
||||
|
||||
log::info!("[input-classifier] → classify_input returned {:?}", result);
|
||||
result
|
||||
}
|
||||
|
||||
async fn classify_input(
|
||||
@@ -96,18 +101,12 @@ impl InputClassifier for HeuristicClassifier {
|
||||
async fn natural_language_detection_heuristic(
|
||||
input: ParsedTokensSnapshot,
|
||||
word_tokens: Vec<String>,
|
||||
current_input_type: InputType,
|
||||
_current_input_type: InputType,
|
||||
include_last_token: bool,
|
||||
) -> ClassificationResult {
|
||||
let word_tokens_count = word_tokens.len();
|
||||
let _word_tokens_count = word_tokens.len();
|
||||
|
||||
let min_token_length = if matches!(current_input_type, InputType::AI) {
|
||||
MINIMUM_COMMAND_DETECTION_TOKEN_LENGTH
|
||||
} else {
|
||||
MINIMUM_NATURAL_LANGUAGE_DETECTION_TOKEN_LENGTH
|
||||
};
|
||||
|
||||
if min_token_length > word_tokens_count as u8 {
|
||||
if input.buffer_text.len() < MINIMUM_CLASSIFICATION_CHAR_LENGTH {
|
||||
return ClassificationResult::pure_shell();
|
||||
}
|
||||
|
||||
|
||||
@@ -58,20 +58,16 @@ pub fn is_prefix_of_natural_language_word(input: &str) -> bool {
|
||||
|
||||
pub async fn is_likely_shell_command(
|
||||
input: &ParsedTokensSnapshot,
|
||||
word_tokens_count: usize,
|
||||
_word_tokens_count: usize,
|
||||
) -> bool {
|
||||
const YIELD_BATCH_SIZE: usize = 5;
|
||||
|
||||
let mut likely_command_token_count = 0;
|
||||
let total_token_count = input.parsed_tokens.len();
|
||||
let mut is_first_token_command = false;
|
||||
for (idx, token) in input.parsed_tokens.iter().enumerate() {
|
||||
// Periodically, yield to the executor so this task can be aborted if
|
||||
// requested.
|
||||
if idx % YIELD_BATCH_SIZE == 0 {
|
||||
futures_lite::future::yield_now().await;
|
||||
}
|
||||
// Early return if we encounter a one-off command / keyword at the beginning of the line.
|
||||
if token.token_index == 0 && ONE_OFF_SHELL_COMMAND_KEYWORDS.contains(&token.token.as_str())
|
||||
{
|
||||
return true;
|
||||
@@ -82,10 +78,6 @@ pub async fn is_likely_shell_command(
|
||||
{
|
||||
likely_command_token_count += 1;
|
||||
}
|
||||
|
||||
if token.token_index == 0 {
|
||||
is_first_token_command = token.token_description.is_some();
|
||||
}
|
||||
}
|
||||
|
||||
// When token count is lower than 2, we should make sure all tokens
|
||||
@@ -98,16 +90,17 @@ pub async fn is_likely_shell_command(
|
||||
DETECT_AS_COMMAND_THRESHOLD
|
||||
};
|
||||
|
||||
// Classify as shell if:
|
||||
// 1) We hit significant threshold of likely shell command tokens.
|
||||
// 2) When there are fewer than 3 tokens, the first token is a valid top-level command.
|
||||
if likely_command_token_count >= (total_token_count as f32 * command_threshold) as usize
|
||||
|| (word_tokens_count < 3 && is_first_token_command)
|
||||
{
|
||||
return true;
|
||||
}
|
||||
|
||||
false
|
||||
let threshold_count = (total_token_count as f32 * command_threshold) as usize;
|
||||
let is_shell = likely_command_token_count >= threshold_count;
|
||||
log::info!(
|
||||
"[input-classifier] is_likely_shell_command: tokens={}, cmd_tokens={}, threshold={:.2} (need {}), result={}",
|
||||
total_token_count,
|
||||
likely_command_token_count,
|
||||
command_threshold,
|
||||
threshold_count,
|
||||
is_shell
|
||||
);
|
||||
is_shell
|
||||
}
|
||||
|
||||
/// Returns true if the first token is a command that is installed on the system.
|
||||
|
||||
Reference in New Issue
Block a user