@skill-graph/cli 0.5.6

package/examples/evals/database-migration.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "database-migration",
+  "subject": "Safe database migration design for expand-contract rollout, locks, backfills, deploy ordering, rollback, and verification",
+  "adjacent_concepts": ["data-modeling", "testing-strategy", "debugging", "owasp-security"],
+  "grounding_note": "Truth sources cite the whole SKILL.md file to keep the initial eval surface stable while routing boundaries are tightened.",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "A migration adds a NOT NULL column to a large production table. Which skill owns the rollout sequence and lock-risk analysis?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/database-migration/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "The schema change needs an expand-contract path, a backfill, application deploy ordering, and rollback conditions. What should database-migration produce?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/database-migration/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "The team is still deciding entity relationships and normalized shape before any DDL is drafted. Should database-migration accept or hand off?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/database-migration/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "A production query is slow and the user asks for profiling and caching strategy, not schema rollout. Which boundary should database-migration respect?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/database-migration/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/debugging.json

@@ -0,0 +1,118 @@
+{
+  "skill_name": "debugging",
+  "subject": "Failure-driven diagnosis workflow: reproduction, scope reduction, evidence capture, root-cause isolation, fix verification, and regression-test creation",
+  "adjacent_concepts": ["testing-strategy", "refactor"],
+  "grounding_note": "This skill is deliberately self-grounding. The debugging workflow is a synthesis of industry practice rather than a derivation from a single canonical text, so truth_sources cite skills/debugging/SKILL.md line ranges rather than external references. The `checkEvalTruthSourceRanges` lint check (D2 in scripts/skill-lint.js) catches citation drift by verifying that every cited line range still exists in the file. Attach external truth_sources only when the skill is later narrowed to a specific methodology (e.g., the SRE debugging runbook).",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "A bug report says \"checkout sometimes fails.\" An engineer is tempted to start reading the checkout code. According to the debugging skill's Workflow table, what is step 1 and why must it come before any hypothesis?",
+      "dimension": "definition",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/debugging/SKILL.md:82-89"]
+    },
+    {
+      "id": 2,
+      "prompt": "The debugging skill's Philosophy section warns that \"a clever fix that skips the reproduction step ships the same bug again under a different name.\" Explain the causal chain behind that claim — why does skipping reproduction make the same failure likely to reappear?",
+      "dimension": "mental_model",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/debugging/SKILL.md:74-76"]
+    },
+    {
+      "id": 3,
+      "prompt": "A team wants to plan test coverage for a newly built billing module that has not yet failed in production. Should they reach for the debugging skill or a different one? Cite the negative-routing rule that decides this.",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/debugging/SKILL.md:109-115"]
+    },
+    {
+      "id": 4,
+      "prompt": "An engineer has a working hypothesis that explains why the error is visible but not why a related field is simultaneously empty. Per the debugging skill's step 4, can they proceed to step 5? If not, what do they do instead and why?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "workflow",
+      "criticality": "critical",
+      "truth_sources": ["skills/debugging/SKILL.md:82-89"]
+    },
+    {
+      "id": 5,
+      "prompt": "The debugging skill requires that a regression test both fail without the fix and pass with it. Why is \"passes with the fix\" alone insufficient, and what failure mode does the two-way check prevent?",
+      "dimension": "purpose",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/debugging/SKILL.md:82-89", "skills/debugging/SKILL.md:101-107"]
+    },
+    {
+      "id": 6,
+      "prompt": "An engineer has spent 90 minutes trying to reproduce a production intermittent failure on their laptop without success. According to the debugging skill's \"When to stop and escalate\" guidance, what does this symptom suggest and what is the correct next move?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/debugging/SKILL.md:91-95"]
+    },
+    {
+      "id": 7,
+      "prompt": "An engineer opens a PR that renames three variables, inlines a helper, and reorders parameter lists inside a module whose test suite is entirely green. No failure has been reported. The engineer asks the debugging skill to review it as a debugging task. Should the skill accept or refuse, and which `Do NOT Use When` row decides this?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/debugging/SKILL.md:109-115"]
+    },
+    {
+      "id": 8,
+      "prompt": "A bug reproduces only when the full application runs — the engineer cannot isolate it to a single module or test. According to Workflow step 2 (Scope), what is the prescribed move, and why is halving preferable to sequentially reading every file in the suspected path?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/debugging/SKILL.md:82-89"]
+    },
+    {
+      "id": 9,
+      "prompt": "A fix has been written and verified against the original failure path (step 5 passes). The engineer is now at step 6. Describe the concrete authoring procedure for the regression test — what input, what assertion, what two-way run — that would satisfy step 6's \"fails without the fix and passes with it\" gate. Distinguish this from merely asserting the fix is present.",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/debugging/SKILL.md:82-89", "skills/debugging/SKILL.md:101-107"]
+    },
+    {
+      "id": 10,
+      "prompt": "An on-call engineer is debugging a production cart-abandonment spike. They notice a null dereference in the checkout path, wrap it in a `try { ... } catch { return null }`, deploy the patch, and the error rate drops. Two weeks later, a different symptom appears (orders with empty line items) that traces back to the same swallowed null. According to the debugging skill's Philosophy and Workflow step 4, is the original patch a valid fix? Name the specific failure mode this commits, explain why the cart-abandonment rate dropping is not evidence the fix was correct, and describe what step 4 required instead.",
+      "dimension": "anti_pattern",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "critical",
+      "truth_sources": ["skills/debugging/SKILL.md:74-76", "skills/debugging/SKILL.md:82-89"]
+    }
+  ]
+}

package/examples/evals/dependency-architecture.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "dependency-architecture",
+  "subject": "Dependency graph architecture for package boundaries, direct and transitive dependencies, adapters, duplicate-purpose libraries, lock-in, upgrades, and supply-chain risk",
+  "adjacent_concepts": ["framework-fit-analysis", "owasp-security", "refactor", "architecture-decision-records"],
+  "grounding_note": "Truth sources cite the whole SKILL.md file to keep the initial eval surface stable while routing boundaries are tightened.",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "A monorepo imports provider SDKs directly from many packages and the team wants dependency boundary rules. Which skill owns the architecture review?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/dependency-architecture/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "The repo has two libraries that solve the same problem and one should be wrapped behind an adapter. What should dependency-architecture inspect?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/dependency-architecture/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "The team has not chosen whether to use Next.js, Remix, or Astro for a new app. Should dependency-architecture accept the framework selection task?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/dependency-architecture/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "The request is a vulnerability-only scan of installed packages. Which boundary should dependency-architecture respect?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/dependency-architecture/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/design-system-architecture.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "design-system-architecture",
+  "subject": "Design-system architecture for token taxonomy, semantic tokens, component APIs, variants, theming, accessibility contracts, governance, and migration",
+  "adjacent_concepts": ["a11y", "microcopy", "layout-composition", "visual-design-foundations"],
+  "grounding_note": "Truth sources cite the whole SKILL.md file to keep the initial eval surface stable while routing boundaries are tightened.",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "A product needs semantic tokens for statuses, chart colors, surfaces, and dark mode before teams keep hardcoding raw colors. Which skill owns this?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/design-system-architecture/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "A component library has inconsistent props, missing state contracts, and no migration path for deprecated tokens. What should design-system-architecture verify?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/design-system-architecture/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "The task is to decide page hierarchy, sitemap, and navigation labels for a product. Should design-system-architecture accept or hand off?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/design-system-architecture/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "The task is to rewrite button labels and empty-state copy, not component APIs or tokens. Which boundary should design-system-architecture respect?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/design-system-architecture/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/error-tracking.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "error-tracking",
+  "subject": "Exception-reporting pipeline design for error boundaries, centralized tracker wrappers, environment gating, PII sanitization, breadcrumbs, and capture verification",
+  "adjacent_concepts": ["observability-modeling", "a11y", "debugging", "owasp-security"],
+  "grounding_note": "Truth sources cite the whole SKILL.md file to keep the initial eval surface stable while routing boundaries are tightened.",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "A React app needs component, route, global, and manual exception reporting paths wired through one sanitized wrapper. Which skill owns the design?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/error-tracking/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "The tracker is receiving customer emails in payloads. According to error-tracking, what must happen before any external send?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/error-tracking/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "The user needs accessible error-message copy and recovery UI for a 404 page. Should error-tracking accept the task?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/error-tracking/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "An observed production exception already exists in the tracker and the task is to find its root cause. Which boundary should error-tracking respect?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/error-tracking/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/event-contract-design.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "event-contract-design",
+  "subject": "Asynchronous event contract design for schemas, envelopes, topics, compatibility, idempotency, ordering, replay, and observability",
+  "adjacent_concepts": ["event-storming", "system-interface-contracts", "webhook-integration", "api-design"],
+  "grounding_note": "Truth sources cite the whole SKILL.md file to keep the initial eval surface stable while the new skill settles.",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "The team has selected OrderPaid as a domain event and now needs topic name, envelope fields, payload schema, idempotency, and versioning. Which skill owns this?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/event-contract-design/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "A team is still discovering commands, policies, aggregates, and the timeline for a business process. Should event-contract-design take over?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/event-contract-design/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "An outbound webhook product sends customer-facing events, but there are no compatibility rules or negative fixtures. Which verification gates fail?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/event-contract-design/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "A provider sends inbound webhook deliveries and the handler must verify HMAC signatures on raw bytes. Which skill owns that mechanics work?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/event-contract-design/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/form-ux-architecture.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "form-ux-architecture",
+  "subject": "Form structure and validation UX across field grouping, validation timing, submission lifecycle, recovery, and client/server split",
+  "adjacent_concepts": ["a11y", "microcopy", "api-design", "data-modeling"],
+  "grounding_note": "Truth sources cite the whole SKILL.md file to keep the initial eval surface stable while the new skill settles.",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "A checkout form asks for 24 fields before shipping can be estimated. How should form-ux-architecture decide what stays, defers, or becomes progressive disclosure?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/form-ux-architecture/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "A form validates every keystroke and shows red errors while the user is still typing. Which validation-timing rule should be applied?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/form-ux-architecture/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "A developer says client-side validation is enough because the browser blocks malformed input. What boundary or philosophy point should form-ux-architecture enforce?",
+      "dimension": "mental_model",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "critical",
+      "truth_sources": ["skills/form-ux-architecture/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "The team asks for final inline validation text after the form lifecycle is settled. Which skill owns that wording?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/form-ux-architecture/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/framework-fit-analysis.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "framework-fit-analysis",
+  "subject": "Technology fit analysis for choosing, replacing, or justifying frameworks, libraries, SDKs, runtimes, databases, UI kits, and platforms",
+  "adjacent_concepts": ["dependency-architecture", "architecture-decision-records", "performance-engineering", "owasp-security"],
+  "grounding_note": "Truth sources cite the whole SKILL.md file to keep the initial eval surface stable while routing boundaries are tightened.",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "A team must choose between Next.js route handlers, server actions, and a separate API service under real deployment constraints. Which skill owns the fit analysis?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/framework-fit-analysis/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "Before adding a charting library, the team wants to weigh ecosystem maturity, migration cost, bundle impact, lock-in, and exit cost. What should framework-fit-analysis evaluate?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/framework-fit-analysis/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "The framework was already chosen and the task is routine dependency hygiene for duplicate packages. Should framework-fit-analysis accept?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/framework-fit-analysis/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "The team has already decided and now needs the formal ADR. Which boundary should framework-fit-analysis respect?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/framework-fit-analysis/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/graph-audit.json

@@ -0,0 +1,139 @@
+{
+  "skill_name": "graph-audit",
+  "subject": "Skill metadata and manifest consistency auditing for the Skill Graph repository",
+  "adjacent_concepts": ["documentation", "refactor", "testing-strategy"],
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "A contributor adds a new skill and sets `eval_artifacts: present` in the frontmatter, but does not create a corresponding eval artifact under `examples/evals/`. What does the graph-audit skill say will happen when `node scripts/skill-lint.js` runs, and which specific check catches this?",
+      "dimension": "definition",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/graph-audit/SKILL.md", "scripts/skill-lint.js"]
+    },
+    {
+      "id": 2,
+      "prompt": "A skill with `scope: codebase` is missing its `grounding` block entirely. According to the graph-audit skill, which two enforcement layers catch this violation — name them and explain the role of each.",
+      "dimension": "mental_model",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/graph-audit/SKILL.md", "schemas/skill.schema.json"]
+    },
+    {
+      "id": 3,
+      "prompt": "When should an agent use the graph-audit skill instead of the refactor skill to address a problem found in a SKILL.md file? Cite the negative-routing boundary that decides this.",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/graph-audit/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "A skill names `relations.verify_with: [missing-skill]` where `missing-skill` does not exist as a directory under `skills/`. What does the graph-audit skill say is the correct fix, and what is the category of this failure mode?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/graph-audit/SKILL.md", "scripts/skill-lint.js"]
+    },
+    {
+      "id": 5,
+      "prompt": "Why does the graph-audit skill set `grounding.grounding_mode: repo_specific` rather than `universal`? What concrete truth sources does this ground it to, and what would be lost if those files changed?",
+      "dimension": "purpose",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/graph-audit/SKILL.md", "docs/skill-metadata-protocol.md"]
+    },
+    {
+      "id": 6,
+      "prompt": "A contributor renames `skills/shopify/` to `skills/shopify-integration/` but forgets to update the `name:` field inside `skills/shopify-integration/SKILL.md` (which still reads `name: shopify`). According to the graph-audit skill, what does `scripts/skill-lint.js` do with this skill, why does this matter for SKILL.md compatibility, and what is the correct fix?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/graph-audit/SKILL.md", "scripts/skill-lint.js", "docs/skill-metadata-protocol.md"]
+    },
+    {
+      "id": 7,
+      "prompt": "An agent regenerates `examples/skills.manifest.sample.json` but uses a different `generated_at` timestamp each time, so the sample drifts from whatever `scripts/generate-manifest.js` would produce on the next run. How does the graph-audit skill say this drift is caught, and why does manifest-sample drift matter even when no real consumer reads the sample?",
+      "dimension": "mental_model",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/graph-audit/SKILL.md", "scripts/generate-manifest.js", "examples/skills.manifest.sample.json"]
+    },
+    {
+      "id": 8,
+      "prompt": "Given the following synthetic SKILL.md frontmatter excerpt:\n\n```yaml\nschema_version: 3\nname: example\ntype: capability\nbrowse_category: quality\nscope: portable\nowner: maintainer\nfreshness: \"2026-04-19\"\ndrift_check:\n  last_verified: \"2026-04-19\"\neval_artifacts: pending\neval_state: passing\nrouting_eval: absent\n```\n\nPredict which check in `scripts/skill-lint.js` will fire first when this skill is linted, what the exact error category will be (per the eleven-check list), and what the underlying rule is. (Hint: one required field has a value that is not in the schema's allowed enum.)",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "output_correctness",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/graph-audit/SKILL.md", "schemas/skill.schema.json", "scripts/skill-lint.js"]
+    },
+    {
+      "id": 9,
+      "prompt": "A skill ships with `schema_version: 2` while the current schema locks `schema_version: const 3`. According to the graph-audit skill, which of the eleven lint checks detects this specifically, and why does `additionalProperties: false` in the schema matter for catching schema drift beyond just the version field?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/graph-audit/SKILL.md", "schemas/skill.schema.json", "scripts/skill-lint.js"]
+    },
+    {
+      "id": 10,
+      "prompt": "A contributor runs `node scripts/skill-lint.js skills/a11y` on a correct skill and receives exit code 0, then introduces a typo changing `eval_artifacts: present` to `eval_artifacts: presnt` (misspelled enum value) and re-runs the same command. According to the graph-audit skill's `## Verification` gate, describe the end-to-end behavioral outcome: what exit code returns, what output category surfaces, which specific check produces the error, and what decision should the contributor's CI make based on the non-zero exit? This is a behavioral assertion — the answer is an observable CI outcome, not a skill-quotation.",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "output_correctness",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/graph-audit/SKILL.md", "scripts/skill-lint.js", "schemas/skill.schema.json"]
+    },
+    {
+      "id": 11,
+      "prompt": "A contributor asks graph-audit to debug why their agent is looping infinitely when it tries to resolve a relation target. Should graph-audit take this task? Cite the `Do NOT Use When` row that decides this and name the correct skill.",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/graph-audit/SKILL.md"]
+    },
+    {
+      "id": 12,
+      "prompt": "A contributor asks graph-audit to help rewrite the `## Coverage` prose in an existing SKILL.md to read more naturally — no schema change, no manifest change, just prose polish. Should graph-audit take this task? Cite the `Do NOT Use When` row that decides this and name the correct skill.",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/graph-audit/SKILL.md"]
+    }
+  ]
+}