yao-meta-skill/reports/output-risk-profile.json

{
  "skill_name": "yao-meta-skill",
  "risk_families": [
    {
      "key": "markdown_readability",
      "label": "Markdown readability",
      "matched_keywords": [
        "md",
        "report",
        "doc"
      ],
      "score": 3,
      "risks": [
        "Tables can render as dense grids with weak hierarchy or poor mobile readability.",
        "Long bullets can make the output look complete while hiding the actual decision logic.",
        "Mixed heading levels can reduce scanability."
      ],
      "constraints": [
        "Use tables only when comparison is the main job; otherwise prefer compact cards or grouped bullets.",
        "Keep table cells short and move explanations below the table.",
        "Use heading levels consistently and keep each section anchored to a user-facing outcome."
      ],
      "self_repair": [
        "Preview whether each table still reads well when columns are narrow.",
        "Convert any table with paragraph-length cells into bullets or cards."
      ]
    },
    {
      "key": "citation_clutter",
      "label": "Citation and footnote clutter",
      "matched_keywords": [
        "source",
        "reference"
      ],
      "score": 2,
      "risks": [
        "Footnote markers or dense citation notes can interrupt the reading flow.",
        "Evidence can be over-attached to obvious statements and under-attached to risky claims.",
        "Source notes may become more prominent than the tutorial itself."
      ],
      "constraints": [
        "Attach citations only to claims that need evidence, not to every sentence.",
        "Group source notes at the end of a section when inline markers would hurt readability.",
        "Keep the main sentence readable without requiring the reader to chase a footnote."
      ],
      "self_repair": [
        "Remove decorative citations that do not support a material claim.",
        "Move repeated source explanations into one compact source note."
      ]
    },
    {
      "key": "visual_capture",
      "label": "Screenshot and visual capture",
      "matched_keywords": [
        "capture"
      ],
      "score": 1,
      "risks": [
        "Screenshots can be captured from the wrong state, wrong viewport, or wrong crop.",
        "Missing screenshots can cause the skill to invent visual references instead of declaring the gap.",
        "Image descriptions can omit the action-relevant region."
      ],
      "constraints": [
        "Never invent a screenshot; state when visual evidence is missing.",
        "Record the source, viewport, and crop intent for any screenshot-dependent output.",
        "Describe what the reader should inspect in the image, not just that an image exists."
      ],
      "self_repair": [
        "Check that every screenshot reference points to a real provided or generated asset.",
        "Reword any visual instruction that depends on an unseen screen state."
      ]
    },
    {
      "key": "code_or_command_safety",
      "label": "Code and command safety",
      "matched_keywords": [
        "script"
      ],
      "score": 1,
      "risks": [
        "Commands can omit environment assumptions, working directory, or rollback notes.",
        "Code snippets can look runnable while missing required inputs.",
        "Error handling can be either absent or over-engineered."
      ],
      "constraints": [
        "Name the working directory, required inputs, and expected output for each command.",
        "Mark destructive or external side-effect operations explicitly.",
        "Prefer the smallest runnable snippet over broad framework code."
      ],
      "self_repair": [
        "Scan each command for cwd, input, output, and side-effect assumptions.",
        "Remove speculative error handling that is not tied to a real failure mode."
      ]
    },
    {
      "key": "tone_and_specificity",
      "label": "Tone and specificity",
      "matched_keywords": [
        "summary"
      ],
      "score": 1,
      "risks": [
        "Headings and summaries can drift into generic, interchangeable language.",
        "The output can sound polished but lose the user's actual taste, audience, or scenario.",
        "Strong claims can appear without examples or constraints."
      ],
      "constraints": [
        "Anchor titles and summaries in the user's audience, object, and concrete outcome.",
        "Avoid placeholder phrases such as comprehensive guide, ultimate solution, or key insights unless the source demands them.",
        "Preserve one distinctive phrase, constraint, or standard from the user's input."
      ],
      "self_repair": [
        "Replace generic title candidates with scenario-specific alternatives.",
        "Delete any polished sentence that could fit almost any project unchanged."
      ]
    }
  ],
  "top_risks": [
    "Tables can render as dense grids with weak hierarchy or poor mobile readability.",
    "Long bullets can make the output look complete while hiding the actual decision logic.",
    "Footnote markers or dense citation notes can interrupt the reading flow.",
    "Evidence can be over-attached to obvious statements and under-attached to risky claims.",
    "Screenshots can be captured from the wrong state, wrong viewport, or wrong crop.",
    "Missing screenshots can cause the skill to invent visual references instead of declaring the gap."
  ],
  "output_constraints": [
    "Use tables only when comparison is the main job; otherwise prefer compact cards or grouped bullets.",
    "Keep table cells short and move explanations below the table.",
    "Attach citations only to claims that need evidence, not to every sentence.",
    "Group source notes at the end of a section when inline markers would hurt readability.",
    "Never invent a screenshot; state when visual evidence is missing.",
    "Record the source, viewport, and crop intent for any screenshot-dependent output."
  ],
  "self_repair_checks": [
    "Preview whether each table still reads well when columns are narrow.",
    "Convert any table with paragraph-length cells into bullets or cards.",
    "Remove decorative citations that do not support a material claim.",
    "Move repeated source explanations into one compact source note.",
    "Check that every screenshot reference points to a real provided or generated asset.",
    "Reword any visual instruction that depends on an unseen screen state."
  ],
  "reviewer_note": "Use this report before deepening the package and again before approving example outputs."
}