Skip to content

/archive-batch skill referenced in ralph, pipeline, seed, and tasks but never implemented (v0.8.0) #29

@graffhyrum

Description

@graffhyrum

Summary

/archive-batch is referenced as the final step of the processing pipeline in at least six source files in v0.8.0, but the skill itself was never implemented. After a batch completes, every entry point that tells the user what to do next points to a command that does not exist.

Affected Files (v0.8.0)

File Reference
skill-sources/ralph/SKILL.md:527 Final Report template: {if batch complete}: Run /archive-batch {batch-id}
skill-sources/pipeline/SKILL.md:47,174,262,283 Phase 4 label and invocation instructions
skill-sources/seed/SKILL.md Phase 2 documentation
skill-sources/tasks/SKILL.md Maintenance task list
platforms/shared/skill-blocks/pipeline.md:59,186,274,283 Phase 5 spec (the most detailed definition)
platforms/shared/skill-blocks/ralph.md:541 Final report next-step hint

Expected Behavior

After /ralph completes a batch with all tasks at status: done, /archive-batch {batch_id} should:

  1. Verify all tasks for the batch are done (guard against partial runs)
  2. Move task files from ops/queue/ to ops/queue/archive/{date}-{batch_id}/
  3. Generate a batch summary file {batch_id}-summary.md in the archive directory
  4. Remove archived batch entries from queue.json

This spec is documented in platforms/shared/skill-blocks/pipeline.md Phase 5 — the implementation just never shipped.

Actual Behavior

Running /archive-batch harvest-skill-html-forms-references (or any batch id) after a completed batch produces:

Unknown skill: archive-batch

The skill-sources/ directory contains no archive-batch/ subdirectory.

Impact

  • Every pipeline run ends with an unexecutable next-step hint
  • Completed batches accumulate in ops/queue/ indefinitely — the queue grows without bound across sessions
  • Users cannot distinguish "in-progress" from "done but not yet archived" batches
  • The pipeline.md Phase 5 spec is dead documentation, creating misleading expectations

Discovery Method

Noticed the dead reference in the ralph final report after running /ralph 11 on a completed batch. Investigation confirmed the reference is in the upstream source (not a local customization): the local skill file is a verbatim copy of the upstream template.

A methodology note in the plugin (methodology/) explicitly lists /archive-batch as a candidate for "skill budget consolidation," suggesting it was planned but deferred before the v0.8.0 release without removing the documentation references.

Workaround

Implemented the skill locally based on the pipeline.md Phase 5 spec. The implementation handles:

  • Completeness verification (with --force escape hatch for broken batches)
  • --dry-run mode
  • Atomic ordering: move files → update queue (not the reverse)
  • Archive directory collision handling for re-runs
  • Missing task file tolerance

Happy to contribute the implementation back if that would be useful.

Environment

  • arscontexta version: 0.8.0
  • Claude Code CLI

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions