Skip to content

Agent stops after ability search while removing category cover images #2697

Description

@superdav42

Description

Two current-version feedback reports show the agent receiving a user request to remove all category cover images, then stopping after two ability-search calls without taking action or returning a final answer. Report 76 was self-reported as worker terminated; report 75 was self-reported as unknown.

Expected Behavior

The agent should continue from ability discovery to a safe implementation plan, use the appropriate term/meta abilities or report a concrete blocker, and return a final user-visible result instead of ending after ability search.

Steps to Reproduce

  1. On a multisite Listeo site, ask: assist in removing all category cover images.
  2. Observe the agent calling wpab__sd-ai-agent__ability-search with remove category cover images taxonomy term images.
  3. Observe the agent calling wpab__sd-ai-agent__ability-search with term meta update delete category image.
  4. The session ends with no final response and no image removal attempt; the feedback system reports unknown or worker terminated.

Environment

  • Plugin version: v1.23.0
  • WordPress: 7.1
  • PHP: 8.3.33
  • Multisite: true
  • Provider: sd-ai-agent-cloud / superdav-chat-pro
  • Theme: listeo
  • Active plugins: not included in compact transcript

Worker Guidance

Inspect the agent loop/runtime path that handles consecutive successful tool calls followed by termination before a final assistant message. Start with the session/agent-loop completion detection and any cloud worker termination mapping used by self-reported feedback. Verify that successful ability-search responses do not leave the session in a terminal state without either another model turn, a tool action, report-inability, or a final answer.

Suggested verification:

  • Add or run a regression covering successful tool responses followed by continued model execution/finalization.
  • Run pnpm run test:php -- --filter=AgentLoop or the narrowest existing session-loop test, then the required full verification before PR.

Feedback Report

Report IDs: 75 (submitted 2026-09-08 09:46:37), 76 (submitted 2026-09-08 10:10:45)


aidevops.sh v3.32.317 plugin for OpenCode v1.18.29 with gpt-5.5 spent 3m and 64,869 tokens on this with the user in an interactive session.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    auto-dispatchAuto-created from TODO.md tagbugAuto-created from TODO.md tagorigin:workerAuto-created by pulse labelless backfill (t2112)status:availableTask is available for claimingtier:standardAuto-created by pulse labelless backfill (t2112)

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions