You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: extensions/copilot/package.nls.json
+1Lines changed: 1 addition & 0 deletions
Original file line number
Diff line number
Diff line change
@@ -482,6 +482,7 @@
482
482
"github.copilot.config.searchSubagent.useAgenticProxy": "Use the agentic proxy for the search subagent tool.",
483
483
"github.copilot.config.searchSubagent.model": "Model to use for the search subagent. When useAgenticProxy is enabled, defaults to 'agentic-search-v3'. Otherwise defaults to the main agent model.",
484
484
"github.copilot.config.searchSubagent.toolCallLimit": "Maximum number of tool calls the search subagent can make during exploration.",
485
+
"github.copilot.config.searchSubagent.thoroughnessEnabled": "Enable the thoroughness parameter on the search subagent tool. When enabled, the caller can pass 'quick', 'medium', or 'thorough' to adjust the number of allowed tool-call turns (0.5×, 1×, or 2× the base toolCallLimit respectively).",
"copilot.tools.executionSubagent.description": "Launch an execution-focused subagent that runs one or more terminal commands to accomplish a task. It is designed to select an efficient summary of the terminal outputs to return to the main agent context.",
487
488
"github.copilot.config.executionSubagent.enabled": "Enable the Execution Subagent tool in Copilot Chat. The Execution Subagent is designed to run terminal commands to accomplish an execution-based task.",
quick: 'Prioritize speed: parallelize as many tool calls as possible, stop as soon as you have a reasonable answer, and avoid deep file reads unless strictly necessary.',
20
+
medium: 'Use a balanced approach: parallelize where possible and stop when you have sufficient context.',
21
+
thorough: 'Be exhaustive: explore all plausible leads, read deeper into files, and verify findings before concluding.',
22
+
};
23
+
16
24
/**
17
25
* Prompt for the search subagent that uses custom search instructions
18
26
* instead of the default agent system prompt.
@@ -28,12 +36,15 @@ export class SearchSubagentPrompt extends PromptElement<SearchSubagentPromptProp
You are an AI coding research assistant that uses search tools to gather information. You can call tools to search for information and read files across a codebase.<br/>
35
45
<br/>
36
-
Once you have thoroughly searched the repository, return a message with ONLY: the <final_answer> tag to provide paths and line ranges of relevant code snippets.<br/>
Once you have searched the repository, return a message with ONLY: the <final_answer> tag to provide paths and line ranges of relevant code snippets.<br/>
37
48
<br/>
38
49
Example:<br/>
39
50
<br/>
@@ -59,4 +70,4 @@ export class SearchSubagentPrompt extends PromptElement<SearchSubagentPromptProp
description: 'Controls the search thoroughness and turn limit. \'quick\' uses fewer turns (fast scan), \'medium\' is the default, \'thorough\' uses more turns for an exhaustive search.',
0 commit comments