Design standalone review follow-up composition

[nothingnesses]

This issue records a proposal. Implementation requires a selected work brief and separate execution approval.

Background and problem

The standalone review prompt works without workflow state. These delivery-role prompts assume a selected work-file action:

• Triager.
• Fixer.
• Verifier.

The library trial required manual adaptations for follow-up roles. Adoption does not authorise changes to those contracts.

Proposal

Compare compact standalone follow-up prompts with explicitly parameterised delivery-role prompts.

Define how a human supplies these inputs without a fictitious selected delivery:

• Reviewed target.
• Criteria.
• Actual findings.
• Permitted correction scope.

Scope and exclusions

Keep this separate from the adoption prompt and configuration proposal.

Prompt composition cannot enforce:

• Safe launchers.
• Isolation.
• Credential protection.
• Protected execution accounting.

Do not present it as a controller.

Acceptance criteria

• The design identifies the authority and inputs required for each follow-up.
• Read-only roles remain separate from correction authority.
• Findings cannot expand the accepted scope.
• Results use direct responses without persistent review state.
• Missing authority returns to the human.
• Any change to existing role contracts requires separate approval.

Alternatives and open questions

Separate prompts preserve current contracts but add assets. Parameterisation reduces duplication but broadens those contracts.

Prefer a small standalone proposal first under Keep changes small and reviewable and No silent scope expansion.

References

• Standalone review prompt.
• triager.
• fixer.
• verifier.
• trial adaptations.