Merge pull request #3225 from code-yeongyu/fix/agent-name-safety

fix(agent): narrow ULW auto-start + strip ZWSP on background path + fix auto-update identity

src/features/background-agent/spawner.test.ts

@@ -466,4 +466,58 @@ describe("background-agent spawner fallback model promotion", () => {
     })
     expect(promptCalls[0]?.body?.variant).toBe("medium")
   })
+
+  test("strips leading zwsp from prompt body agent before promptAsync", async () => {
+    //#given
+    const promptCalls: Array<{ body?: { agent?: string } }> = []
+
+    const client = {
+      session: {
+        get: async () => ({ data: { directory: "/parent/dir" } }),
+        create: async () => ({ data: { id: "ses_child_clean_agent" } }),
+        promptAsync: async (args?: { body?: { agent?: string } }) => {
+          promptCalls.push(args ?? {})
+          return {}
+        },
+      },
+    }
+
+    const task = createTask({
+      description: "Test task",
+      prompt: "Do work",
+      agent: "\u200Bsisyphus-junior",
+      parentSessionID: "ses_parent",
+      parentMessageID: "msg_parent",
+    })
+
+    const item = {
+      task,
+      input: {
+        description: task.description,
+        prompt: task.prompt,
+        agent: task.agent,
+        parentSessionID: task.parentSessionID,
+        parentMessageID: task.parentMessageID,
+        parentModel: task.parentModel,
+        parentAgent: task.parentAgent,
+        model: task.model,
+      },
+    }
+
+    const ctx = {
+      client,
+      directory: "/fallback",
+      concurrencyManager: { release: () => {} },
+      tmuxEnabled: false,
+      onTaskError: () => {},
+    }
+
+    //#when
+    await startTask(item as any, ctx as any)
+    await new Promise((resolve) => setTimeout(resolve, 0))
+
+    //#then
+    expect(promptCalls).toHaveLength(1)
+    expect(promptCalls[0]?.body?.agent).toBe("sisyphus-junior")
+  })
 })

src/features/background-agent/spawner.ts

@@ -6,6 +6,7 @@ import { applySessionPromptParams } from "../../shared/session-prompt-params-hel
 import { subagentSessions } from "../claude-code-session-state"
 import { getTaskToastManager } from "../task-toast-manager"
 import { isInsideTmux } from "../../shared/tmux"
+import { stripAgentListSortPrefix } from "../../shared/agent-display-names"
 import type { ConcurrencyManager } from "./concurrency"
 
 export const FALLBACK_AGENT = "general"
@@ -168,11 +169,12 @@ export async function startTask(
       }
     : undefined
   const launchVariant = input.model?.variant
+  const normalizedAgent = stripAgentListSortPrefix(input.agent)
 
   applySessionPromptParams(sessionID, input.model)
 
   const promptBody = {
-    agent: input.agent,
+    agent: normalizedAgent,
     ...(launchModel ? { model: launchModel } : {}),
     ...(launchVariant ? { variant: launchVariant } : {}),
     system: input.skillContent,
@@ -180,7 +182,7 @@ export async function startTask(
       task: false,
       call_omo_agent: true,
       question: false,
-      ...getAgentToolRestrictions(input.agent),
+      ...getAgentToolRestrictions(normalizedAgent),
     },
     parts: [createInternalAgentTextPart(input.prompt)],
   }

src/hooks/auto-update-checker/checker/plugin-entry.test.ts

@@ -4,6 +4,7 @@ import * as fs from "node:fs"
 import * as os from "node:os"
 import * as path from "node:path"
 import { PACKAGE_NAME } from "../constants"
+import { LEGACY_PLUGIN_NAME, PLUGIN_NAME } from "../../../shared/plugin-identity"
 
 type PluginEntryResult = {
   entry: string
@@ -120,6 +121,64 @@ describe("findPluginEntry", () => {
     expect(pluginInfo?.pinnedVersion).toBe("3.5.2")
   })
 
+  test("finds preferred plugin entry", async () => {
+    // #given preferred plugin entry is configured
+    fs.writeFileSync(configPath, JSON.stringify({ plugin: [PLUGIN_NAME] }))
+
+    // #when plugin entry is detected
+    const execution = runFindPluginEntry(temporaryDirectory)
+
+    // #then preferred entry is returned
+    expect(execution.status).toBe(0)
+    const pluginInfo = JSON.parse(execution.stdout.trim()) as PluginEntryResult
+    expect(pluginInfo?.entry).toBe(PLUGIN_NAME)
+    expect(pluginInfo?.isPinned).toBe(false)
+    expect(pluginInfo?.pinnedVersion).toBeNull()
+  })
+
+  test("finds legacy plugin entry", async () => {
+    // #given legacy plugin entry is configured
+    fs.writeFileSync(configPath, JSON.stringify({ plugin: [LEGACY_PLUGIN_NAME] }))
+
+    // #when plugin entry is detected
+    const execution = runFindPluginEntry(temporaryDirectory)
+
+    // #then legacy entry is returned
+    expect(execution.status).toBe(0)
+    const pluginInfo = JSON.parse(execution.stdout.trim()) as PluginEntryResult
+    expect(pluginInfo?.entry).toBe(LEGACY_PLUGIN_NAME)
+    expect(pluginInfo?.isPinned).toBe(false)
+    expect(pluginInfo?.pinnedVersion).toBeNull()
+  })
+
+  test("finds preferred plugin entry with pinned version", async () => {
+    // #given preferred plugin entry includes semver version
+    fs.writeFileSync(configPath, JSON.stringify({ plugin: [`${PLUGIN_NAME}@3.15.0`] }))
+
+    // #when plugin entry is detected
+    const execution = runFindPluginEntry(temporaryDirectory)
+
+    // #then preferred versioned entry is returned
+    expect(execution.status).toBe(0)
+    const pluginInfo = JSON.parse(execution.stdout.trim()) as PluginEntryResult
+    expect(pluginInfo?.entry).toBe(`${PLUGIN_NAME}@3.15.0`)
+    expect(pluginInfo?.isPinned).toBe(true)
+    expect(pluginInfo?.pinnedVersion).toBe("3.15.0")
+  })
+
+  test("returns null for unrelated plugin entry", async () => {
+    // #given unrelated plugin entry is configured
+    fs.writeFileSync(configPath, JSON.stringify({ plugin: ["some-other-plugin"] }))
+
+    // #when plugin entry is detected
+    const execution = runFindPluginEntry(temporaryDirectory)
+
+    // #then no matching entry is returned
+    expect(execution.status).toBe(0)
+    const pluginInfo = JSON.parse(execution.stdout.trim()) as PluginEntryResult
+    expect(pluginInfo).toBeNull()
+  })
+
   test("reads user config from profile dir even when OPENCODE_CONFIG_DIR changes after import", async () => {
     // #given profile-specific user config after module import
     const profileConfigDir = path.join(temporaryDirectory, "profiles", "today")

src/hooks/auto-update-checker/checker/plugin-entry.ts

@@ -3,6 +3,7 @@ import type { OpencodeConfig } from "../types"
 import { PACKAGE_NAME } from "../constants"
 import { getConfigPaths } from "./config-paths"
 import { stripJsonComments } from "./jsonc-strip"
+import { LEGACY_PLUGIN_NAME, PLUGIN_NAME } from "../../../shared/plugin-identity"
 
 export interface PluginEntryInfo {
   entry: string
@@ -12,6 +13,7 @@ export interface PluginEntryInfo {
 }
 
 const EXACT_SEMVER_REGEX = /^\d+\.\d+\.\d+(-[0-9A-Za-z-]+(\.[0-9A-Za-z-]+)*)?(\+[0-9A-Za-z-]+(\.[0-9A-Za-z-]+)*)?$/
+const MATCH_PLUGIN_NAMES = [PACKAGE_NAME, PLUGIN_NAME, LEGACY_PLUGIN_NAME]
 
 export function findPluginEntry(directory: string): PluginEntryInfo | null {
   for (const configPath of getConfigPaths(directory)) {
@@ -22,13 +24,15 @@ export function findPluginEntry(directory: string): PluginEntryInfo | null {
       const plugins = config.plugin ?? []
 
       for (const entry of plugins) {
-        if (entry === PACKAGE_NAME) {
-          return { entry, isPinned: false, pinnedVersion: null, configPath }
-        }
-        if (entry.startsWith(`${PACKAGE_NAME}@`)) {
-          const pinnedVersion = entry.slice(PACKAGE_NAME.length + 1)
-          const isPinned = EXACT_SEMVER_REGEX.test(pinnedVersion.trim())
-          return { entry, isPinned, pinnedVersion, configPath }
+        for (const pluginName of MATCH_PLUGIN_NAMES) {
+          if (entry === pluginName) {
+            return { entry, isPinned: false, pinnedVersion: null, configPath }
+          }
+          if (entry.startsWith(`${pluginName}@`)) {
+            const pinnedVersion = entry.slice(pluginName.length + 1)
+            const isPinned = EXACT_SEMVER_REGEX.test(pinnedVersion.trim())
+            return { entry, isPinned, pinnedVersion, configPath }
+          }
         }
       }
     } catch {

src/hooks/keyword-detector/hook-ralph-loop.test.ts

@@ -86,6 +86,44 @@ describe("keyword-detector ralph-loop activation", () => {
     expect(startLoopCalls[0].options.ultrawork).toBe(true)
   })
 
+  test("#given ulw mentioned mid-sentence #when chat.message fires #then ralph-loop startLoop is not invoked", async () => {
+    // given
+    setMainSession("main-session")
+    const startLoopCalls: StartLoopCall[] = []
+    const ralphLoop = createMockRalphLoop(startLoopCalls)
+    const hook = createKeywordDetectorHook(createMockPluginInput(), undefined, ralphLoop)
+    const output = {
+      message: {} as Record<string, unknown>,
+      parts: [{ type: "text", text: "I think ulw is cool" }],
+    }
+
+    // when
+    await hook["chat.message"]({ sessionID: "main-session", agent: "sisyphus" }, output)
+
+    // then
+    expect(startLoopCalls).toHaveLength(0)
+    expect(output.parts[0]?.text).toBe("I think ulw is cool")
+  })
+
+  test("#given question about ultrawork #when chat.message fires #then ralph-loop startLoop is not invoked", async () => {
+    // given
+    setMainSession("main-session")
+    const startLoopCalls: StartLoopCall[] = []
+    const ralphLoop = createMockRalphLoop(startLoopCalls)
+    const hook = createKeywordDetectorHook(createMockPluginInput(), undefined, ralphLoop)
+    const output = {
+      message: {} as Record<string, unknown>,
+      parts: [{ type: "text", text: "what is ultrawork?" }],
+    }
+
+    // when
+    await hook["chat.message"]({ sessionID: "main-session", agent: "sisyphus" }, output)
+
+    // then
+    expect(startLoopCalls).toHaveLength(0)
+    expect(output.parts[0]?.text).toBe("what is ultrawork?")
+  })
+
   test("#given non-ulw message #when chat.message fires #then ralph-loop startLoop is not invoked", async () => {
     // given
     setMainSession("main-session")

src/hooks/keyword-detector/hook.ts

@@ -16,11 +16,16 @@ import type { RalphLoopHook } from "../ralph-loop"
 import { parseRalphLoopArguments } from "../ralph-loop/command-arguments"
 
 const ULTRAWORK_KEYWORD_PATTERN = /\b(ultrawork|ulw)\b/i
+const LEADING_ULTRAWORK_PATTERN = /^\s*(ultrawork|ulw)\b/i
 
 function extractUltraworkTask(cleanText: string): string {
   return cleanText.replace(ULTRAWORK_KEYWORD_PATTERN, "").trim()
 }
 
+function hasLeadingUltraworkKeyword(cleanText: string): boolean {
+  return LEADING_ULTRAWORK_PATTERN.test(cleanText)
+}
+
 export function createKeywordDetectorHook(
   ctx: PluginInput,
   _collector?: ContextCollector,
@@ -76,6 +81,16 @@ export function createKeywordDetectorHook(
         }
       }
 
+      if (!hasLeadingUltraworkKeyword(cleanText)) {
+        const preFilterCount = detectedKeywords.length
+        detectedKeywords = detectedKeywords.filter((k) => k.type !== "ultrawork")
+        if (preFilterCount > detectedKeywords.length) {
+          log(`[keyword-detector] Filtered non-leading ultrawork keyword`, {
+            sessionID: input.sessionID,
+          })
+        }
+      }
+
       if (detectedKeywords.length === 0) {
         return
       }

src/tools/delegate-task/background-task.test.ts

@@ -204,6 +204,50 @@ describeFn("executeBackgroundTask output/session metadata compatibility", () =>
     ])
   })
 
+  testFn("strips leading zwsp from agent name before launching background task", async () => {
+    //#given - display-sorted agent names should be normalized before manager launch
+    const launchCalls: unknown[] = []
+    const manager = {
+      launch: async (input: unknown) => {
+        launchCalls.push(input)
+        return {
+          id: "bg_clean_agent",
+          sessionID: "ses_clean_agent",
+          description: "Clean agent",
+          agent: "sisyphus-junior",
+          status: "running",
+        }
+      },
+      getTask: () => ({ sessionID: "ses_clean_agent" }),
+    }
+
+    //#when
+    await executeBackgroundTask(
+      {
+        description: "Clean agent",
+        prompt: "check",
+        run_in_background: true,
+        load_skills: [],
+      },
+      {
+        sessionID: "ses_parent",
+        callID: "call_clean_agent",
+        metadata: async () => {},
+        abort: new AbortController().signal,
+      },
+      { manager },
+      { sessionID: "ses_parent", messageID: "msg_clean_agent" },
+      "\u200Bsisyphus-junior",
+      undefined,
+      undefined,
+      undefined,
+    )
+
+    //#then
+    expectFn(launchCalls).toHaveLength(1)
+    expectFn((launchCalls[0] as { agent: string }).agent).toBe("sisyphus-junior")
+  })
+
   testFn("keeps launched background task alive when parent aborts before session id resolves", async () => {
     //#given - parallel tool execution can abort the parent call after launch succeeds
     const metadataCalls: any[] = []

src/tools/delegate-task/background-task.ts

@@ -10,6 +10,7 @@ import { getSessionTools } from "../../shared/session-tools-store"
 import { SessionCategoryRegistry } from "../../shared/session-category-registry"
 import { QUESTION_DENIED_SESSION_PERMISSION } from "../../shared/question-denied-session-permission"
 import { setSessionFallbackChain } from "../../hooks/model-fallback/hook"
+import { stripAgentListSortPrefix } from "../../shared/agent-display-names"
 
 function continueSessionSetup(args: {
   taskID: string
@@ -62,11 +63,12 @@ export async function executeBackgroundTask(
 
   try {
     const tddEnabled = executorCtx.sisyphusAgentConfig?.tdd
-    const effectivePrompt = buildTaskPrompt(args.prompt, agentToUse, tddEnabled)
+    const normalizedAgent = stripAgentListSortPrefix(agentToUse)
+    const effectivePrompt = buildTaskPrompt(args.prompt, normalizedAgent, tddEnabled)
     const task = await manager.launch({
       description: args.description,
       prompt: effectivePrompt,
-      agent: agentToUse,
+      agent: normalizedAgent,
       parentSessionID: parentContext.sessionID,
       parentMessageID: parentContext.messageID,
       parentModel: parentContext.model,
@@ -156,7 +158,7 @@ Do NOT call background_output now. Wait for <system-reminder> notification first
     return formatDetailedError(error, {
       operation: "Launch background task",
       args,
-      agent: agentToUse,
+      agent: stripAgentListSortPrefix(agentToUse),
       category: args.category,
     })
   }