specfuse

review-scenario

"Request an independent pre-merge review of an Arazzo scenario or recipe -- spawns the scenario-reviewer subagent, which evaluates the file cold (without the original PM intent) across seven quality and granularity categories and returns an APPROVE / REQUEST CHANGES / NEEDS DISCUSSION verdict. Use before merging a scenario for an unbiased second opinion."

specfuse 0 Updated 3mo ago
GitHub

Install

npx skillscat add specfuse/authoring/review-scenario

Install via the SkillsCat registry.

SKILL.md

Request an independent pre-merge review of an Arazzo scenario or recipe file. Spawns the scenario-reviewer sub-agent, which evaluates the file cold -- without seeing the original PM intent or design conversation.

Enforces: .specfuse/authoring/handbooks/Arazzo_Handbook.md

Input

Accept the target file. The user may provide:

  • A full file path (e.g., api/specs/v1/domains/order/scenarios/order-lifecycle.arazzo.yaml)
  • A domain and scenario name (e.g., "order order-lifecycle")

If ambiguous, list matching files and ask the user to pick.

Process

  1. Read the target file. Confirm it exists and is a valid .arazzo.yaml or .recipe.yaml file.

  2. Gather domain context. Read the same-domain corpus for granularity comparison:

    • Glob api/specs/v1/domains/{domain}/scenarios/*.arazzo.yaml for all existing scenarios in the target's domain. Extract workflowIds and step operationId sequences from each.
    • Glob api/specs/v1/scenarios/setup-recipes/**/*.recipe.yaml for all existing recipes. Extract file stems and output keys.
  3. Spawn the scenario-reviewer subagent (provided by the specfuse-authoring plugin). Use the Agent tool with a prompt that:

    • Relies on the subagent's built-in review process
    • Provides the target file path
    • Provides the existing scenario list with workflowIds and step operationId sequences
    • Provides the existing recipe list with file stems and output keys
    • Instructs the agent to follow the 7-category review process (A through G) and return the structured report format

    Independence constraint: Do NOT pass any of the following to the reviewer agent:

    • The original PM conversation or use-case description
    • The /design-scenario session history
    • The architect agent's generation output or reasoning
    • Any commentary about why the scenario was designed this way

    The reviewer must form its own judgment from the YAML alone.

  4. Present the review report. Show the structured output to the user:

    • Verdict: APPROVE, REQUEST CHANGES, or NEEDS DISCUSSION
    • Checklist with per-category status
    • Granularity assessment (if overlap detected)
    • Findings ordered by severity
    • Recommendation summary
  5. Handle the verdict:

    • APPROVE: Report that the scenario is ready to merge. Note any warnings for optional follow-up.
    • REQUEST CHANGES: List each error finding with its suggested action. The author should address these before merging.
    • NEEDS DISCUSSION: Highlight the granularity assessment. The human reviewer must decide whether the overlap is acceptable or the scenarios should be consolidated.

Checklist

  • Target file identified and read
  • Domain context gathered (existing scenarios and recipes)
  • Scenario-reviewer agent spawned with file and context only (no PM intent)
  • Structured review report presented to the user
  • Verdict and recommended actions clearly communicated