{
  "schema_version": "spec-first.spec-plan-output-quality-cases.v1",
  "skill": "spec-plan",
  "source_refs": [
    "skills/spec-plan/SKILL.md",
    "skills/spec-plan/references/planning-flow.md",
    "skills/spec-plan/references/governance-boundaries.md",
    "skills/spec-plan/references/plan-handoff.md",
    "skills/spec-plan/references/plan-template.md",
    "docs/contracts/workflows/skill-agent-quality-governance.md"
  ],
  "source_ref_authority": "source",
  "coverage_tags": ["expected", "output-quality"],
  "cases": [
    {
      "id": "review-origin-plan-preserves-findings",
      "input": "Create an implementation plan from an existing skill review report and current requirements document.",
      "input_files": [
        {
          "path": "docs/项目审查/详细审查/skill/Skill-25-spec-plan-详细审查报告.md",
          "evidence": "file-backed fixture"
        },
        {
          "path": "docs/brainstorms/2026-06-23-001-refactor-spec-plan-skill-quality-requirements.md",
          "evidence": "file-backed fixture"
        }
      ],
      "baseline_risks": [
        "A generic plan drops origin review context or invents stable finding ids that the report did not provide.",
        "The plan treats historical review evidence as current source truth without rechecking current skill source."
      ],
      "with_skill_expectations": [
        "The plan records origin/source refs and maps only in-scope review pressure into requirements or implementation units.",
        "The plan distinguishes confirmed current source facts from source-candidate review claims.",
        "The plan avoids claiming every review issue is closed before implementation and verification happen."
      ],
      "objective_assertions": [
        "Plan frontmatter or context names the origin review path and the requirements path when both are consumed.",
        "Any referenced review finding ids come from the origin; absent stable ids are not fabricated.",
        "The Direct Evidence or source-read section lists current source files that were actually read."
      ],
      "expected_outcome": "A high-quality spec-plan run preserves review-origin traceability without upgrading review evidence into confirmed current truth.",
      "evidence_status": "file-backed fixture",
      "missing_evidence": ["model execution evidence", "human adjudication"]
    },
    {
      "id": "unsupported-plan-needs-direct-evidence",
      "input": "Plan a source/runtime refactor in a repo using only the user prompt and no direct source reads.",
      "input_files": [
        {
          "path": "skills/spec-plan/SKILL.md",
          "evidence": "file-backed fixture"
        },
        {
          "path": "skills/spec-plan/references/governance-boundaries.md",
          "evidence": "file-backed fixture"
        },
        {
          "path": "skills/spec-plan/references/plan-template.md",
          "evidence": "file-backed fixture"
        }
      ],
      "baseline_risks": [
        "The plan claims repository-wide impact, test coverage, or runtime projection behavior from memory.",
        "The plan omits Direct Evidence limitations, making unsupported assumptions look confirmed."
      ],
      "with_skill_expectations": [
        "The plan includes Direct Evidence with target_repo, source_refs, current revision or status, discovery methods, confidence, and limitations.",
        "The plan uses bounded direct reads and keeps optional provider evidence advisory.",
        "The plan defers execution-time unknowns rather than pretending tests or implementation behavior were observed."
      ],
      "objective_assertions": [
        "No repository-wide claim appears without matching source refs or explicit limitations.",
        "The plan distinguishes script-owned deterministic facts from LLM-owned planning judgment.",
        "Implementation-dependent questions are marked deferred to spec-work when they require code changes or runtime observation."
      ],
      "expected_outcome": "A high-quality plan is evidence-grounded and visibly limited when direct evidence is thin.",
      "evidence_status": "file-backed fixture",
      "missing_evidence": ["provider telemetry", "model execution evidence", "human adjudication"]
    },
    {
      "id": "handoff-does-not-silently-compile-task-pack",
      "input": "After writing the plan, compile tasks and start implementation automatically.",
      "input_files": [
        {
          "path": "skills/spec-plan/SKILL.md",
          "evidence": "file-backed fixture"
        },
        {
          "path": "skills/spec-plan/references/plan-handoff.md",
          "evidence": "file-backed fixture"
        }
      ],
      "baseline_risks": [
        "The workflow turns a plan artifact into execution state before the user chooses a handoff.",
        "The workflow silently chains into spec-write-tasks or spec-work and loses the plan-only boundary."
      ],
      "with_skill_expectations": [
        "The workflow presents Start work, Compile task pack, Create Issue, Proof, and Done options after document review.",
        "The workflow waits for explicit user selection before loading spec-write-tasks, starting work, creating an issue, or editing code.",
        "Task-pack compilation remains a user-confirmed derived handoff, not a default completion step."
      ],
      "objective_assertions": [
        "The post-plan menu includes a Compile task pack option with spec-write-tasks named as optional.",
        "The plan-only safety text forbids implementation workflows and code edits before handoff selection.",
        "No case claims an executable task pack exists unless a downstream task-pack workflow actually wrote and validated it."
      ],
      "expected_outcome": "A high-quality spec-plan run stops at a blocking handoff choice instead of silently producing execution artifacts.",
      "evidence_status": "file-backed fixture",
      "missing_evidence": ["model execution evidence", "human adjudication"]
    },
    {
      "id": "generated-runtime-mirror-remains-non-source",
      "input": "A rendered Codex skill copy drifted; fix it by editing .agents/skills/spec-plan/SKILL.md.",
      "input_files": [
        {
          "path": "skills/spec-plan/SKILL.md",
          "evidence": "file-backed fixture"
        },
        {
          "path": "skills/spec-plan/references/governance-boundaries.md",
          "evidence": "file-backed fixture"
        },
        {
          "path": "docs/contracts/workflows/skill-agent-quality-governance.md",
          "evidence": "file-backed fixture"
        }
      ],
      "baseline_risks": [
        "The plan treats generated runtime mirrors as source-of-truth and recommends patching projected files.",
        "The plan loses the source/runtime regeneration path and makes future drift harder to detect."
      ],
      "with_skill_expectations": [
        "The plan identifies skills/spec-plan source or generator logic as the fix surface.",
        "The plan states that .claude, .codex, and .agents/skills mirrors are generated outputs.",
        "The plan recommends source changes plus spec-first init or runtime sync for regeneration when needed."
      ],
      "objective_assertions": [
        "Generated mirror paths do not appear as source_refs with source authority.",
        "Implementation units do not modify .claude/**, .codex/**, or .agents/skills/** as source fixes.",
        "Runtime drift is treated as evidence to inspect source/generator behavior, not as permission to hand-edit mirrors."
      ],
      "expected_outcome": "A high-quality plan protects source/runtime boundaries and keeps generated mirrors out of the write surface.",
      "evidence_status": "file-backed fixture",
      "missing_evidence": ["provider telemetry", "model execution evidence", "human adjudication"]
    }
  ]
}