Skip to content

Show workspace provenance per trial in capability digests #103

Description

@Jordak

Context

PR #102 adds base-only workspaces and records provenance in result JSON and trial/reference reports: original task repository, pinned task commit, workspace history policy, and synthetic workspace base ref.

A review-pass follow-up found that capability digest views still summarize workspace base ref at run-context level. In mixed-task contexts that can collapse into a mixed: ... value without mapping each synthetic base ref back to the trial/task that used it.

Desired Behavior

Capability digests should expose enough per-trial or per-task provenance to audit which task repo/commit and synthetic workspace base ref produced each row, without making the main summary unreadable.

Suggested Scope

  • Update Markdown capability digests to include per-trial or per-task provenance for:
    • task repository
    • task commit
    • workspace base ref
  • Update HTML capability digests with equivalent provenance.
  • Keep workspace history policy in the context summary if useful.
  • Avoid making digest tables excessively wide if a clearer detail row/expandable section works better.

Acceptance Criteria

  • A digest with multiple tasks can map every trial row to its task repo, task commit, and workspace base ref.
  • Existing result JSON and trial report provenance behavior remains unchanged.
  • Tests cover a multi-task or mixed-base-ref digest case.

Source

Deferred from PR #102 review-pass as non-blocking IF6 so the base-only workspace implementation PR can stay focused on workspace mechanics.

Design readiness: ready to implement

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

    ready-for-agentFully specified and ready for an AFK agent

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions