Files
ds2api/internal/util/tool_prompt.go

153 lines
5.7 KiB
Go

package util
import "strings"
// BuildToolCallInstructions generates the unified tool-calling instruction block
// used by all adapters (OpenAI, Claude, Gemini). It uses attention-optimized
// structure: rules → negative examples → positive examples → anchor.
//
// The toolNames slice should contain the actual tool names available in the
// current request; the function picks real names for examples.
func BuildToolCallInstructions(toolNames []string) string {
// Pick real tool names for examples; fall back to generic names.
ex1 := "read_file"
ex2 := "write_to_file"
ex3 := "ask_followup_question"
used := map[string]bool{}
for _, n := range toolNames {
switch {
// Read/query-type tools
case !used["ex1"] && matchAny(n, "read_file", "list_files", "search_files", "Read", "Glob"):
ex1 = n
used["ex1"] = true
// Write/execute-type tools
case !used["ex2"] && matchAny(n, "write_to_file", "apply_diff", "execute_command", "exec_command", "Write", "Edit", "MultiEdit", "Bash"):
ex2 = n
used["ex2"] = true
// Interactive/meta tools
case !used["ex3"] && matchAny(n, "ask_followup_question", "attempt_completion", "update_todo_list", "Task"):
ex3 = n
used["ex3"] = true
}
}
ex1Params := exampleReadParams(ex1)
ex2Params := exampleWriteOrExecParams(ex2)
ex3Params := exampleInteractiveParams(ex3)
return `TOOL CALL FORMAT — FOLLOW EXACTLY:
When calling tools, the final block of your response MUST be raw XML. Do not use markdown fences. Do not put any text after the XML block.
<tool_calls>
<tool_call>
<tool_name>TOOL_NAME_HERE</tool_name>
<parameters>{"key":"value"}</parameters>
</tool_call>
</tool_calls>
RULES:
1) When calling tools, the final block of your response MUST be the XML above. Any explanatory text must appear before that block, never after it.
2) <parameters> MUST contain a strict JSON object. All JSON keys and strings use double quotes.
3) Multiple tools → multiple <tool_call> blocks inside ONE <tool_calls> root.
4) Do NOT wrap the XML in markdown code fences (no triple backticks).
5) After receiving a tool result, use it directly. Only call another tool if the result is insufficient.
6) If you want to say something AND call a tool, output text first, then the XML block on its own. Do not narrate pseudo-tool calls in plain text.
7) Parameters MUST use the exact field names from the selected tool schema.
8) CRITICAL: Do NOT invent or add any extra fields (such as "_raw", "_xml"). Use ONLY the fields strictly defined in the schema. Extra fields will cause execution failure.
9) Never claim that a tool was run, or that you saw search/browser/file results, unless you actually received a tool result for that tool call.
10) If a tool call fails, is rejected, or returns no usable result: do NOT fabricate the missing result. Either retry once with corrected parameters when the fix is obvious, or ask the user / answer with the limitation.
11) Never output internal tool-planning traces such as "let me search", "calling tool", "<tool_calls>", "<tool_call>", or fake tool transcripts unless you are emitting a real tool call block at the end.
❌ WRONG — Do NOT do these:
Wrong 1 — mixed text and XML:
I'll read the file for you. <tool_calls><tool_call>...
Wrong 2 — describing tool calls in text:
[调用 Bash] {"command": "ls"}
Wrong 3 — missing <tool_calls> wrapper:
<tool_call><tool_name>` + ex1 + `</tool_name><parameters>{}</parameters></tool_call>
Wrong 4 — extra/invented fields:
<parameters>{"_raw": "...", "command": "ls"}</parameters>
Wrong 5 — fake tool execution/result narration:
Let me search the web first.
Search results show the answer is 42.
Wrong 6 — tool failed, but model pretends it succeeded:
The tool returned three skills, so install review-implementing.
✅ CORRECT EXAMPLES:
Example A — Single tool:
<tool_calls>
<tool_call>
<tool_name>` + ex1 + `</tool_name>
<parameters>` + ex1Params + `</parameters>
</tool_call>
</tool_calls>
Example B — Two tools in parallel:
<tool_calls>
<tool_call>
<tool_name>` + ex1 + `</tool_name>
<parameters>` + ex1Params + `</parameters>
</tool_call>
<tool_call>
<tool_name>` + ex2 + `</tool_name>
<parameters>` + ex2Params + `</parameters>
</tool_call>
</tool_calls>
Example C — Tool with complex nested JSON parameters:
<tool_calls>
<tool_call>
<tool_name>` + ex3 + `</tool_name>
<parameters>` + ex3Params + `</parameters>
</tool_call>
</tool_calls>
Remember: when calling tools, end with ONE real <tool_calls>...</tool_calls> XML block and do not invent tool results.`
}
func matchAny(name string, candidates ...string) bool {
for _, c := range candidates {
if name == c {
return true
}
}
return false
}
func exampleReadParams(name string) string {
switch strings.TrimSpace(name) {
case "Read":
return `{"file_path":"README.md"}`
case "Glob":
return `{"pattern":"**/*.go","path":"."}`
default:
return `{"path":"src/main.go"}`
}
}
func exampleWriteOrExecParams(name string) string {
switch strings.TrimSpace(name) {
case "Bash", "execute_command":
return `{"command":"pwd"}`
case "exec_command":
return `{"cmd":"pwd"}`
case "Edit":
return `{"file_path":"README.md","old_string":"foo","new_string":"bar"}`
case "MultiEdit":
return `{"file_path":"README.md","edits":[{"old_string":"foo","new_string":"bar"}]}`
default:
return `{"path":"output.txt","content":"Hello world"}`
}
}
func exampleInteractiveParams(name string) string {
switch strings.TrimSpace(name) {
case "Task":
return `{"description":"Investigate flaky tests","prompt":"Run targeted tests and summarize failures"}`
default:
return `{"question":"Which approach do you prefer?","follow_up":[{"text":"Option A"},{"text":"Option B"}]}`
}
}