Skip to content

Open-source LLM knowledge platform: turn raw documents into a queryable RAG, an autonomous reasoning agent, and a self-maintaining Wiki.

fix(agent): keep skill-script stdout on failure for the model and UI

wizardchenf79052fe673e
diff --git a/frontend/src/composables/useChatStreamHandler.test.mjs b/frontend/src/composables/useChatStreamHandler.test.mjs
new file mode 100644
index 000000000..afc21ecf2
--- /dev/null
+++ b/frontend/src/composables/useChatStreamHandler.test.mjs
@@ -0,0 +1,13 @@
+import assert from 'node:assert/strict'
+import { readFileSync } from 'node:fs'
+import test from 'node:test'
+
+const source = readFileSync(new URL('./useChatStreamHandler.ts', import.meta.url), 'utf8')
+
+test('failed tool results keep stdout/output instead of replacing it with the short error', () => {
+ assert.match(source, /toolCallEvent\.output = dataPayload\.output \|\| data\.content/)
+ assert.doesNotMatch(
+ source,
+ /toolCallEvent\.output = success\s*\?[\s\S]*dataPayload\.error/,
+ )
+})
diff --git a/frontend/src/composables/useChatStreamHandler.ts b/frontend/src/composables/useChatStreamHandler.ts
index 4f903175a..71ec58c7c 100644
--- a/frontend/src/composables/useChatStreamHandler.ts
+++ b/frontend/src/composables/useChatStreamHandler.ts
@@ -760,9 +760,10 @@ export function useChatStreamHandler(options: UseChatStreamHandlerOptions) {
if (toolCallEvent) {
toolCallEvent.pending = false
toolCallEvent.success = success
- toolCallEvent.output = success
- ? dataPayload.output || data.content
- : dataPayload.error || data.content
+ // Keep stdout/markdown on failure. The error field is often just
+ // "exited with code 1" plus a retry hint; the streams live on
+ // output / tool_data and are what the terminal card should show.
+ toolCallEvent.output = dataPayload.output || data.content
toolCallEvent.error = !success ? dataPayload.error || data.content : undefined
const duration =
dataPayload.duration_ms !== undefined ? dataPayload.duration_ms : dataPayload.duration
diff --git a/frontend/src/utils/shellExecResult.test.ts b/frontend/src/utils/shellExecResult.test.ts
index 580629593..6183d1613 100644
--- a/frontend/src/utils/shellExecResult.test.ts
+++ b/frontend/src/utils/shellExecResult.test.ts
@@ -48,3 +48,31 @@ test('previewShellCommand collapses whitespace and truncates', () => {
assert.equal(previewShellCommand('ls -la'), 'ls -la')
assert.equal(previewShellCommand('abcdefghij', 8), 'abcdefg…')
})
+
+test('buildShellExecView parses skill-script markdown and command fallback', () => {
+ const view = buildShellExecView(
+ {
+ skill_name: 'smart-charts',
+ script_path: 'scripts/cli.py',
+ args: ['--x-axis', '工作项目'],
+ exit_code: 1,
+ },
+ null,
+ [
+ '=== Script Execution: smart-charts/scripts/cli.py ===',
+ '',
+ '**Arguments**: [--x-axis 工作项目]',
+ '**Exit Code**: 1',
+ '',
+ '## Standard Output',
+ '',
+ '```',
+ '{"chart":{"success":false,"error":{"error":"X轴字段不存在:工作项目"}}}',
+ '```',
+ '',
+ ].join('\n'),
+ )
+ assert.equal(view.command, 'smart-charts/scripts/cli.py --x-axis 工作项目')
+ assert.equal(view.exitCode, 1)
+ assert.match(view.stdout, /X轴字段不存在:工作项目/)
+})
diff --git a/frontend/src/utils/shellExecResult.ts b/frontend/src/utils/shellExecResult.ts
index 07393730c..6d822b7b0 100644
--- a/frontend/src/utils/shellExecResult.ts
+++ b/frontend/src/utils/shellExecResult.ts
@@ -64,7 +64,9 @@ function extractFencedSection(output: string, heading: string): string {
function stripLegacyShellHeader(output: string): string {
return output
.replace(/^=== Shell Exec[\s\S]*?\n\n/, '')
+ .replace(/^=== Script Execution[\s\S]*?\n\n/, '')
.replace(/\*\*Command\*\*:[\s\S]*?(?:\n(?!\*\*)|$)/, '')
+ .replace(/\*\*Arguments\*\*:.*\n/, '')
.replace(/\*\*Work Dir\*\*:.*\n/, '')
.replace(/\*\*Exit Code\*\*:.*\n/, '')
.replace(/\*\*Duration\*\*:.*\n/, '')
@@ -74,6 +76,21 @@ function stripLegacyShellHeader(output: string): string {
.trim()
}
+function skillScriptCommand(
+ data: Record<string, unknown>,
+ args: Record<string, unknown>,
+): string {
+ const skill = asString(data.skill_name) || asString(args.skill_name)
+ const script = asString(data.script_path) || asString(args.script_path)
+ if (!skill && !script) return ''
+ const path = [skill, script].filter(Boolean).join('/')
+ const extra = data.args ?? args.args
+ if (Array.isArray(extra) && extra.length) {
+ return [path, ...extra.map((item) => asString(item))].join(' ')
+ }
+ return path
+}
+
export function buildShellExecView(
data: Record<string, unknown> | null | undefined,
args: unknown,
@@ -84,8 +101,12 @@ export function buildShellExecView(
const rawOutput = asString(output)
const stdoutFromData = asString(record.stdout)
const stderrFromData = asString(record.stderr)
- const parsedStdout = stdoutFromData ? '' : extractFencedSection(rawOutput, 'Stdout')
- const parsedStderr = stderrFromData ? '' : extractFencedSection(rawOutput, 'Stderr')
+ const parsedStdout = stdoutFromData
+ ? ''
+ : extractFencedSection(rawOutput, 'Stdout') || extractFencedSection(rawOutput, 'Standard Output')
+ const parsedStderr = stderrFromData
+ ? ''
+ : extractFencedSection(rawOutput, 'Stderr') || extractFencedSection(rawOutput, 'Standard Error')
let stdout = stdoutFromData || parsedStdout
let stderr = stderrFromData || parsedStderr
@@ -94,7 +115,7 @@ export function buildShellExecView(
}
return {
- command: asString(record.command) || asString(argumentsRecord.command),
+ command: asString(record.command) || asString(argumentsRecord.command) || skillScriptCommand(record, argumentsRecord),
workDir: asString(record.work_dir) || asString(argumentsRecord.work_dir),
exitCode: asNumber(record.exit_code),
durationMs: asNumber(record.duration_ms),
diff --git a/frontend/src/views/chat/components/AgentStreamDisplay.vue b/frontend/src/views/chat/components/AgentStreamDisplay.vue
index 34d2cba07..b6f0c82b3 100644
--- a/frontend/src/views/chat/components/AgentStreamDisplay.vue
+++ b/frontend/src/views/chat/components/AgentStreamDisplay.vue
@@ -1071,7 +1071,9 @@ const isMcpTool = (toolName?: string | null): boolean => String(toolName || '').
const resolveToolDisplayType = (event: any): DisplayType | undefined => {
if (event?.display_type) return event.display_type as DisplayType
- if (event?.tool_name === 'shell_exec') return 'shell_exec'
+ if (event?.tool_name === 'shell_exec' || event?.tool_name === 'execute_skill_script') {
+ return 'shell_exec'
+ }
return undefined
};
@@ -2676,7 +2678,7 @@ const getToolTitle = (event: any): string => {
if (event.tool_name === 'wiki_search' || event.tool_name === 'wiki_read_page') {
return `${getLocalizedToolName(event.tool_name)}...`;
}
- if (event.tool_name === 'shell_exec') {
+ if (event.tool_name === 'shell_exec' || event.tool_name === 'execute_skill_script') {
return t('agentStream.toolStatus.shellExecRunning');
}
const localizedName = getLocalizedToolName(event.tool_name);
@@ -2774,10 +2776,8 @@ const getToolTitle = (event: any): string => {
return pageLabel ? `${baseTitle}:「${sanitizeForDisplay(pageLabel)}」` : baseTitle;
}
- if (toolName === 'shell_exec') {
- const command = previewShellCommand(
- String(event.tool_data?.command || event.arguments?.command || ''),
- )
+ if (toolName === 'shell_exec' || toolName === 'execute_skill_script') {
+ const command = previewShellCommand(skillScriptCommandLabel(event))
const baseTitle = getToolDescription(event)
return command ? `${baseTitle}:${command}` : baseTitle
}
@@ -2787,6 +2787,19 @@ const getToolTitle = (event: any): string => {
return summary || getToolDescription(event);
};
+const skillScriptCommandLabel = (event: any): string => {
+ const fromData = String(event?.tool_data?.command || event?.arguments?.command || '').trim()
+ if (fromData) return fromData
+ const skill = String(event?.tool_data?.skill_name || event?.arguments?.skill_name || '').trim()
+ const script = String(event?.tool_data?.script_path || event?.arguments?.script_path || '').trim()
+ const path = [skill, script].filter(Boolean).join('/')
+ const args = event?.tool_data?.args || event?.arguments?.args
+ if (Array.isArray(args) && args.length) {
+ return [path, ...args.map((arg: unknown) => String(arg))].filter(Boolean).join(' ')
+ }
+ return path
+}
+
// Tool description
const getToolDescription = (event: any): string => {
if (event.pending) {
@@ -2799,7 +2812,7 @@ const getToolDescription = (event: any): string => {
if (event.tool_name === 'query_understand') {
return t('agentStream.toolStatus.queryUnderstanding');
}
- if (event.tool_name === 'shell_exec') {
+ if (event.tool_name === 'shell_exec' || event.tool_name === 'execute_skill_script') {
return t('agentStream.toolStatus.shellExecRunning');
}
const localizedName = getLocalizedToolName(event.tool_name);
@@ -2832,7 +2845,7 @@ const getToolDescription = (event: any): string => {
return success ? t('agentStream.toolStatus.attachmentParsingDone') : t('agentStream.toolStatus.attachmentParsingFailed');
} else if (toolName === 'query_understand') {
return success ? t('agentStream.toolStatus.queryUnderstandDone') : t('agentStream.toolStatus.calledFailed', { name: getLocalizedToolName(toolName) });
- } else if (toolName === 'shell_exec') {
+ } else if (toolName === 'shell_exec' || toolName === 'execute_skill_script') {
const localizedName = getLocalizedToolName(toolName);
return success ? localizedName : t('agentStream.toolStatus.calledFailed', { name: localizedName });
} else {
diff --git a/internal/agent/tools/persist.go b/internal/agent/tools/persist.go
index 149c1e2ca..347d8c421 100644
--- a/internal/agent/tools/persist.go
+++ b/internal/agent/tools/persist.go
@@ -144,17 +144,17 @@ func CompactToolOutputForHistory(toolName string, result *types.ToolResult) stri
if result == nil {
return ""
}
- if !result.Success {
- if result.Error != "" {
- return "Error: " + result.Error
- }
- return "Error: tool call failed"
- }
if isSandboxContentTool(toolName) {
if rebuilt := compactSandboxHistory(result); rebuilt != "" {
- return rebuilt
+ if result.Success {
+ return rebuilt
+ }
+ return failedToolVisibleContent(rebuilt, result.Error)
}
}
+ if !result.Success {
+ return failedToolVisibleContent(result.Output, result.Error)
+ }
if result.Output != "" && !ShouldOmitRawToolOutput(toolName, result.Data) {
return result.Output
}
@@ -162,7 +162,25 @@ func CompactToolOutputForHistory(toolName string, result *types.ToolResult) stri
}
func isSandboxContentTool(toolName string) bool {
- return toolName == ToolShellExec || toolName == ToolReadSandboxFile
+ return toolName == ToolShellExec || toolName == ToolReadSandboxFile || toolName == ToolExecuteSkillScript
+}
+
+// failedToolVisibleContent keeps stdout/stderr (in Output) when a tool fails.
+// Error is often a one-line exit status plus a retry hint; the streams are
+// what the model needs to change arguments instead of guessing.
+func failedToolVisibleContent(output, errMsg string) string {
+ output = strings.TrimSpace(output)
+ errMsg = strings.TrimSpace(errMsg)
+ switch {
+ case output == "" && errMsg == "":
+ return "Error: tool call failed"
+ case output == "":
+ return "Error: " + errMsg
+ case errMsg == "" || strings.Contains(output, errMsg):
+ return output
+ default:
+ return output + "\n\nError: " + errMsg
+ }
}
func compactHistoricalSandboxOutput(output string) string {
diff --git a/internal/agent/tools/persist_test.go b/internal/agent/tools/persist_test.go
index a713f2fd1..53831a854 100644
--- a/internal/agent/tools/persist_test.go
+++ b/internal/agent/tools/persist_test.go
@@ -230,3 +230,57 @@ func TestCompactToolOutputForHistory_recoversStreamsFromPlaceholder(t *testing.T
t.Fatalf("rebuilt history should keep the command, got %q", history)
}
}
+
+func TestCompactToolOutputForHistory_failedSkillScriptKeepsStdout(t *testing.T) {
+ stdout := `{"chart":{"success":false,"error":{"error":"X轴字段不存在:工作项目","available":["name","value"]}}}`
+ output := "=== Script Execution: smart-charts/scripts/cli.py ===\n\n**Exit Code**: 1\n\n## Standard Output\n\n```\n" + stdout + "\n```\n"
+ errMsg := "Script exited with code 1\n\n[Analyze the error above and try a different approach.]"
+ history := CompactToolOutputForHistory(ToolExecuteSkillScript, &types.ToolResult{
+ Success: false,
+ Output: output,
+ Error: errMsg,
+ Data: map[string]interface{}{
+ "display_type": "shell_exec",
+ "command": "smart-charts/scripts/cli.py --x-axis 工作项目",
+ "exit_code": 1,
+ "stdout": stdout,
+ },
+ })
+ if !strings.Contains(history, "X轴字段不存在:工作项目") {
+ t.Fatalf("failed skill script history must keep stdout, got %q", history)
+ }
+ if !strings.Contains(history, "Error: Script exited with code 1") {
+ t.Fatalf("failed skill script history must still surface the error, got %q", history)
+ }
+}
+
+func TestSanitizeAgentStepsForStorage_skillScriptKeepsStreamsOnFailure(t *testing.T) {
+ stdout := `{"chart":{"success":false,"error":{"error":"X轴字段不存在:工作项目"}}}`
+ output := "=== Script Execution: smart-charts/scripts/cli.py ===\n\n" + stdout
+ steps := []types.AgentStep{{
+ ToolCalls: []types.ToolCall{{
+ Name: ToolExecuteSkillScript,
+ Result: &types.ToolResult{
+ Success: false,
+ Output: output,
+ Error: "Script exited with code 1",
+ Data: map[string]interface{}{
+ "display_type": "shell_exec",
+ "command": "smart-charts/scripts/cli.py",
+ "exit_code": 1,
+ "stdout": stdout,
+ "stderr": "",
+ },
+ },
+ }},
+ }}
+
+ sanitized := SanitizeAgentStepsForStorage(steps)
+ result := sanitized[0].ToolCalls[0].Result
+ if strings.Contains(result.Output, "omitted from history") {
+ t.Fatalf("failed skill script must not collapse to an omit line, got %q", result.Output)
+ }
+ if got, _ := result.Data["stdout"].(string); !strings.Contains(got, "X轴字段不存在") {
+ t.Fatalf("persisted stdout should remain for the UI card, got %#v", result.Data["stdout"])
+ }
+}
diff --git a/internal/agent/tools/skill_execute.go b/internal/agent/tools/skill_execute.go
index c80b3de40..aedb5e75d 100644
--- a/internal/agent/tools/skill_execute.go
+++ b/internal/agent/tools/skill_execute.go
@@ -206,14 +206,16 @@ func (t *ExecuteSkillScriptTool) Execute(ctx context.Context, args json.RawMessa
success := result.IsSuccess()
resultData := map[string]interface{}{
- "skill_name": input.SkillName,
- "script_path": input.ScriptPath,
- "args": input.Args,
- "exit_code": result.ExitCode,
- "stdout": result.Stdout,
- "stderr": result.Stderr,
- "duration_ms": result.Duration.Milliseconds(),
- "killed": result.Killed,
+ "display_type": "shell_exec",
+ "command": skillScriptCommand(input),
+ "skill_name": input.SkillName,
+ "script_path": input.ScriptPath,
+ "args": input.Args,
+ "exit_code": result.ExitCode,
+ "stdout": result.Stdout,
+ "stderr": result.Stderr,
+ "duration_ms": result.Duration.Milliseconds(),
+ "killed": result.Killed,
}
logger.Infof(ctx, "[Tool][ExecuteSkillScript] Script completed with exit code: %d", result.ExitCode)
@@ -234,6 +236,18 @@ func (t *ExecuteSkillScriptTool) Execute(ctx context.Context, args json.RawMessa
}, nil
}
+func skillScriptCommand(input ExecuteSkillScriptInput) string {
+ parts := make([]string, 0, 1+len(input.Args))
+ switch {
+ case input.SkillName != "" && input.ScriptPath != "":
+ parts = append(parts, input.SkillName+"/"+input.ScriptPath)
+ case input.ScriptPath != "":
+ parts = append(parts, input.ScriptPath)
+ }
+ parts = append(parts, input.Args...)
+ return strings.Join(parts, " ")
+}
+
// Cleanup releases any resources
func (t *ExecuteSkillScriptTool) Cleanup(ctx context.Context) error {
return nil
diff --git a/internal/application/service/agent_history.go b/internal/application/service/agent_history.go
index a471f0a03..614f21d32 100644
--- a/internal/application/service/agent_history.go
+++ b/internal/application/service/agent_history.go
@@ -210,18 +210,12 @@ func filterNonTerminalToolCalls(calls []types.ToolCall) []types.ToolCall {
}
// toolCallOutput returns the textual content to use for a historical tool
-// message: the recorded Output on success, or an "Error: …" line otherwise so
-// the model can tell that an earlier tool call failed.
+// message. Failures still go through CompactToolOutputForHistory so stdout
+// from a crashed skill script is not dropped in favor of a one-line exit code.
func toolCallOutput(tc types.ToolCall) string {
if tc.Result == nil {
return ""
}
- if !tc.Result.Success {
- if tc.Result.Error != "" {
- return "Error: " + tc.Result.Error
- }
- return "Error: tool call failed"
- }
return agenttools.CompactToolOutputForHistory(tc.Name, tc.Result)
}
diff --git a/internal/application/service/agent_history_test.go b/internal/application/service/agent_history_test.go
index 1f83d7fd6..9630087c6 100644
--- a/internal/application/service/agent_history_test.go
+++ b/internal/application/service/agent_history_test.go
@@ -250,6 +250,42 @@ func TestBuildAssistantHistoryMessages_ToolFailureSurfacesAsError(t *testing.T)
}, got[1])
}
+func TestBuildAssistantHistoryMessages_SkillScriptFailureKeepsStdout(t *testing.T) {
+ stdout := `{"chart":{"success":false,"error":{"error":"X轴字段不存在:工作项目"}}}`
+ msg := &types.Message{
+ Role: "assistant",
+ Content: "I will retry with a different axis.",
+ AgentSteps: types.AgentSteps{
+ {
+ Iteration: 0,
+ Thought: "Plot the chart.",
+ ToolCalls: []types.ToolCall{
+ {
+ ID: "call_skill",
+ Name: agenttools.ToolExecuteSkillScript,
+ Args: map[string]interface{}{"skill_name": "smart-charts", "script_path": "scripts/cli.py"},
+ Result: &types.ToolResult{
+ Success: false,
+ Output: "=== Script Execution ===\n" + stdout,
+ Error: "Script exited with code 1",
+ Data: map[string]interface{}{
+ "display_type": "shell_exec",
+ "stdout": stdout,
+ "exit_code": 1,
+ },
+ },
+ },
+ },
+ },
+ },
+ }
+ got := buildAssistantHistoryMessages(msg)
+ require.Len(t, got, 3)
+ assert.Equal(t, "tool", got[1].Role)
+ assert.Contains(t, got[1].Content, "X轴字段不存在:工作项目")
+ assert.Contains(t, got[1].Content, "Error: Script exited with code 1")
+}
+
// TestFilterNonTerminalToolCalls confirms a legacy final_answer entry is
// dropped — every other tool (KB search, web search, MCP tools…) must survive.
func TestFilterNonTerminalToolCalls(t *testing.T) {
diff --git a/internal/modelcontext/model_output.go b/internal/modelcontext/model_output.go
index 80d358d6c..1d9b6bbd7 100644
--- a/internal/modelcontext/model_output.go
+++ b/internal/modelcontext/model_output.go
@@ -27,10 +27,7 @@ func (r *sourceRegistry) ModelOutput(result *types.ToolResult) string {
return r.modelWebFetchOutput(mapsValue(result.Data["results"]), result.Output)
}
if !result.Success {
- if result.Error != "" {
- return "Error: " + r.CompactKnownText(result.Error)
- }
- return "Error: tool call failed"
+ return failedToolModelText(result.Output, result.Error)
}
switch displayType {
case "grep_results":
@@ -478,6 +475,26 @@ func truncateModelEvidence(value string, maxRunes int) (string, bool) {
return string(runes[:maxRunes]), true
}
+// failedToolModelText is what the model should see for a failed tool call.
+// Script and shell failures put the useful diagnostics in Output (stdout /
+// stderr); Error is often just "exited with code 1" plus a retry hint.
+// Putting Output first makes "[Analyze the error above ...]" refer to the
+// actual streams.
+func failedToolModelText(output, errMsg string) string {
+ output = strings.TrimSpace(output)
+ errMsg = strings.TrimSpace(errMsg)
+ switch {
+ case output == "" && errMsg == "":
+ return "Error: tool call failed"
+ case output == "":
+ return "Error: " + errMsg
+ case errMsg == "" || strings.Contains(output, errMsg):
+ return output
+ default:
+ return output + "\n\nError: " + errMsg
+ }
+}
+
func mapsValue(value interface{}) []map[string]interface{} {
if value == nil {
return nil
diff --git a/internal/modelcontext/registry.go b/internal/modelcontext/registry.go
index 591e1f618..233f70ebe 100644
--- a/internal/modelcontext/registry.go
+++ b/internal/modelcontext/registry.go
@@ -250,7 +250,7 @@ func (r *Registry) ModelToolResultForTool(toolName string, result *types.ToolRes
if result.Success {
return result.Output
}
- return result.Error
+ return failedToolModelText(result.Output, result.Error)
}
// Protect durable resources and UUID-bearing summary slugs before the
// source codec sees any embedded document IDs. Encode once more afterwards
@@ -266,10 +266,8 @@ func (r *Registry) ModelToolResultForTool(toolName string, result *types.ToolRes
modelOutput = r.sources.ModelOutput(&copyResult)
} else if copyResult.Success {
modelOutput = copyResult.Output
- } else if copyResult.Error != "" {
- modelOutput = "Error: " + copyResult.Error
} else {
- modelOutput = "Error: tool call failed"
+ modelOutput = failedToolModelText(copyResult.Output, copyResult.Error)
}
// Even tools without structured source results can surface a known durable
// ID in validation errors or status text. Compact only explicitly declared
diff --git a/internal/modelcontext/registry_test.go b/internal/modelcontext/registry_test.go
index 061dc4b5e..dd7e4f09f 100644
--- a/internal/modelcontext/registry_test.go
+++ b/internal/modelcontext/registry_test.go
@@ -272,6 +272,19 @@ func TestRegistryCompactsKnownIDsInBuiltInValidationErrors(t *testing.T) {
require.Equal(t, "Error: document d1 belongs to knowledge base b1", got)
}
+func TestModelToolResultForTool_failedSkillScriptKeepsStdout(t *testing.T) {
+ registry := NewRegistry(true)
+ stdout := `{"chart":{"success":false,"error":{"error":"X轴字段不存在:工作项目","available":["name","value"]}}}`
+ got := registry.ModelToolResultForTool("execute_skill_script", &types.ToolResult{
+ Success: false,
+ Output: "=== Script Execution: smart-charts/scripts/cli.py ===\n\n## Standard Output\n\n```\n" + stdout + "\n```\n",
+ Error: "Script exited with code 1\n\n[Analyze the error above and try a different approach.]",
+ })
+ require.Contains(t, got, "X轴字段不存在:工作项目")
+ require.Contains(t, got, "available")
+ require.Contains(t, got, "Error: Script exited with code 1")
+}
+
func TestRegistryCompactsDatabaseQueryIDColumnsForBuiltInFollowUps(t *testing.T) {
registry := NewRegistry(true)
modelOutput := registry.ModelToolResultForTool("database_query", &types.ToolResult{