Skip to content

Trivial gaps should offer inline fix, not full plan-set/execute-set ceremony #53

Description

@skytect
Human NoteMissing re-export shouldn't require a plan-set --gaps and execute-set --gaps but that is what is recommended right now
Version6.0.0
Skillexecute-set

Problem Statement

When post-execution verification identifies trivial gaps (e.g., a single missing re-export in a barrel file), the execute-set skill recommends the full gap-closure ceremony: /rapid:plan-set N --gaps followed by /rapid:execute-set N --gaps. This is disproportionate overhead for a one-line fix. The user ends up just doing it manually, which means the gap-closure workflow is effectively bypassed for the most common class of gaps.

The current gap reporting in execute-set always suggests the same next step regardless of gap severity:

To close gaps: /rapid:plan-set {SET_INDEX} --gaps then /rapid:execute-set {SET_INDEX} --gaps

There is no distinction between a gap that requires architectural rework and one that needs a single line change.

Proposed Solution

Classify gaps by complexity during verification and offer tiered resolution paths:

  1. Trivial gaps (missing exports, typos, minor config additions): Offer an inline /rapid:quick-style fix directly from the execute-set skill, or let the verifier agent fix them on the spot before writing GAPS.md.
  2. Non-trivial gaps (missing functionality, behavioral mismatches): Continue recommending the full plan-set --gaps / execute-set --gaps ceremony.

The verifier agent already has enough context to judge severity. It could return a severity: "trivial" | "non-trivial" field per gap in its RAPID:RETURN, and the execute-set orchestrator could act accordingly.

Current Workaround

Fix trivial gaps manually (read the file, make the one-line edit, commit) instead of running the gap-closure workflow.

Related Issues

None identified.

Co-Authored-By: Claude Opus 4.6

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

    ai-authoredIssue was written with AI assistanceenhancementNew feature or request

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions