Back to Openspec

Agent Handoff Output And Delivery Polish Evidence

openspec/initiatives/context-store-and-initiatives/work-items/13-agent-handoff-output-and-delivery-polish/evidence.md

1.8.01.6 KB
Original Source

Agent Handoff Output And Delivery Polish Evidence

Manual Beta Source Notes

The manual beta pass found:

  • after context-store setup, the user is told to ask an agent to create an initiative, but a fresh agent may not know the beta CLI or where to find the playbook;
  • initiative create --json reports created_files as relative names, so agents must combine them with the returned root before writing;
  • "commands only" can sound like "the agent gets no OpenSpec guidance," even though users may only mean slash commands as workflow entrypoints;
  • tools without command adapters need a clear warning when workflow slash commands cannot be installed.

Initial Recommendation

Treat this as output polish, not a new workflow engine:

  • add direct path fields rather than breaking existing relative fields;
  • keep handoff guidance concrete and command-sized;
  • keep baseline OpenSpec literacy separate from workflow entrypoints;
  • leave the broader "what should I do next?" command to the proposed handoff work item.

Reassessment

After reviewing practical examples, the proposed "Next for your agent" shape looked too prescriptive. It assumes a fixed linear beta path, but agents may inspect state, branch, skip setup, continue an existing change, or use context in a different order.

Conclusion on 2026-05-30:

  • Skip Item 13 as an implementation item for now.
  • Preserve the evidence because the handoff pain is real.
  • Do not hardcode fixed next-step guidance until the product has a clearer receipt or affordance model.
  • Consider splitting deterministic created_paths style receipt fields into a smaller future slice if they remain useful.