Skip to content

[extend] Goal mode: same-session autonomous long-horizon loop #48240

Closed as not planned

Description

@johnhenry030888

Opencode has plan and build modes but no goal mode for long-horizon autonomous tasks. Proposal: an event-sourced single-goal domain with CAS revisions, model-facing get/create/update goal tools, an idle-checkpoint same-session round driver (one retained goal-round prompt per admitted round, human input always wins), a /goal command, and a goal status projection. Design follows proven DSH goal-mode semantics (durable phase vs process-local armed activation, round-budget caps, evidence-before-complete, 3-round blocker discipline). Scope: shared session/tool/command plus TUI/desktop UI.

Reference implementation exists locally (same-session loop + CAS revisions + fault-injection tests); happy to shape it into a PR if the direction is welcome.

Activity

  1. github-actions commented on Sep 9, 2026

    @github-actions
    Contributor

    This issue doesn't fully meet our contributing guidelines.

    What needs to be fixed:

    • No recognized issue template was used. Please use one of the three templates: Bug Report, Feature Request (title must start with [FEATURE]: and include the verification checkbox), or Question.
    • The [extend] prefix is not a valid template type. If this is a feature request, please reopen using the Feature Request template with the required verification checkbox and a [FEATURE]: title prefix.

    Please edit this issue to address the above within 2 hours, or it will be automatically closed.


    This issue might be a duplicate of existing issues. Please check:

    If you believe this was flagged incorrectly, please let a maintainer know.

  2. johnhenry030888 commented on Sep 9, 2026

    @johnhenry030888
    Author

    Closing in favor of #27167, which proposes the same /goal command, per-session goal persistence, model-visible goal tools and TUI/desktop support. What this proposal added on top: a working reference design — event-sourced single-goal domain with CAS revisions, durable-phase vs process-local activation, round-budget caps, evidence-before-complete, 3-round blocker discipline, plus a reconcile() that retracts orphan synthetic rounds. If #27167 moves forward, I can shape the reference implementation (same-session round driver + tests) into a PR.

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

Metadata

Metadata

Assignees

Labels

needs:complianceThis means the issue will auto-close after 2 hours.

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions