@skill-graph/cli 0.5.6

package/examples/evals/information-architecture.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "information-architecture",
+  "subject": "Information architecture for navigation, page hierarchy, sitemaps, wayfinding, labeling systems, content grouping, and findability",
+  "adjacent_concepts": ["task-analysis", "taxonomy-design", "layout-composition", "microcopy"],
+  "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 docs site has useful content but users cannot find setup instructions. Which skill owns navigation, grouping, and wayfinding improvements?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/information-architecture/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "A product team is deciding whether information belongs in top-level nav, tabs, filters, or a page section. What should information-architecture inspect?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/information-architecture/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "The task is to build a formal category taxonomy and assignment rules for the skill library. Should information-architecture accept?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/information-architecture/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "The route hierarchy is known and the next task is responsive grid and breakpoint composition. Which boundary should information-architecture respect?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/information-architecture/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/interaction-feedback.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "interaction-feedback",
+  "subject": "UI feedback-state staging for user actions, asynchronous work, success, failure, retry, undo, and perceived latency",
+  "adjacent_concepts": ["microcopy", "a11y", "state-machine-modeling", "performance-engineering"],
+  "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 user clicks Export and nothing changes for 8 seconds before a file appears. Which feedback states should interaction-feedback design before wording is written?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/interaction-feedback/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "A toast message is too vague and needs better copy. Should interaction-feedback own the words?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/interaction-feedback/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "A save action updates the UI optimistically, then the server rejects the change. What must the interaction-feedback verification cover?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/interaction-feedback/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "An order lifecycle has pending, paid, fulfilled, refunded, and cancelled states with legal transitions. Should interaction-feedback define the lifecycle itself?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/interaction-feedback/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/interaction-patterns.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "interaction-patterns",
+  "subject": "Choosing and auditing UI interaction patterns and controls before implementation",
+  "adjacent_concepts": ["a11y", "task-analysis", "interaction-feedback", "design-system-architecture"],
+  "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 user needs to compare five mutually exclusive pricing options before choosing one. Should the interface hide them in a dropdown? Apply the interaction-patterns method.",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/interaction-patterns/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "A custom combobox has already been chosen and now needs ARIA roles, focus behavior, and keyboard support. Which skill owns that next step?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/interaction-patterns/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "A flow uses a modal to edit a record while the user needs to compare values in the table behind it. What should interaction-patterns inspect before recommending inline edit, drawer, or modal?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/interaction-patterns/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "A team asks interaction-patterns to define component props and semantic tokens for the selected control. Should the skill continue or hand off?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/interaction-patterns/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/layout-composition.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "layout-composition",
+  "subject": "Responsive layout and screen composition for page structure, scan pattern, breakpoints, and stable viewport hierarchy",
+  "adjacent_concepts": ["task-analysis", "information-architecture", "visual-design-foundations", "design-system-architecture"],
+  "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 route contract already names primary, secondary, and supporting content. The user asks how to order the sections on desktop and mobile. Which skill owns the next step, and what should it produce?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/layout-composition/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "A dashboard looks fine at 1440px but the chart controls jump above the title and the table overflows on mobile. According to the layout-composition skill, what verification gates are failing?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/layout-composition/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "The user asks for a color palette and type scale after the grid is already settled. Should layout-composition accept the task or route elsewhere?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/layout-composition/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "A card grid changes height when a status label appears, causing nearby buttons to move. Which layout-composition principle applies?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "workflow",
+      "criticality": "normal",
+      "truth_sources": ["skills/layout-composition/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/lint-overlay.json

@@ -0,0 +1,117 @@
+{
+  "skill_name": "lint-overlay",
+  "subject": "Overlay for testing-strategy covering lint as verification: rule selection, gate placement, failure triage, and migration planning when introducing rules to an existing codebase",
+  "adjacent_concepts": ["testing-strategy", "refactor", "debugging"],
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "A contributor loads lint-overlay alone and asks it to decide whether lint belongs in a change's verification plan at all. According to the overlay's Extends section, is that a valid use? If not, what does the overlay say must be loaded alongside it and why?",
+      "dimension": "definition",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "critical",
+      "truth_sources": ["skills/lint-overlay/SKILL.md:65-82"]
+    },
+    {
+      "id": 2,
+      "prompt": "The overlay says \"effort-to-risk matching\" is NOT overridden — the base testing-strategy skill still decides whether to lint before this overlay decides how. Explain why the overlay deliberately leaves that decision to the base rather than re-deriving it.",
+      "dimension": "mental_model",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/lint-overlay/SKILL.md:65-82"]
+    },
+    {
+      "id": 3,
+      "prompt": "An engineer is staring at a single failing ESLint rule in one file and wants help fixing that specific violation. Should they load lint-overlay? Cite the negative-routing rule that decides this and name the skill that should be loaded instead.",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/lint-overlay/SKILL.md:118-124"]
+    },
+    {
+      "id": 4,
+      "prompt": "A team wants to introduce a new `no-console` lint rule to an existing codebase that already has ~200 violations. According to the lint-overlay's Overlay Rules table, what is the correct migration approach and why does adding a rule that immediately fails 200 files count as a scope change rather than a test?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/lint-overlay/SKILL.md:102-112"]
+    },
+    {
+      "id": 5,
+      "prompt": "The lint-overlay places lint after unit tests but before integration tests in the verification sequence. Why this specific ordering? What property of lint failures makes running them earlier than integration tests a win, and what property of unit tests makes them still earlier?",
+      "dimension": "purpose",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/lint-overlay/SKILL.md:102-112"]
+    },
+    {
+      "id": 6,
+      "prompt": "A reviewer argues that lint failures on pre-existing violations should be downgraded to warnings because the current change did not introduce them. According to the lint-overlay, is that correct? Cite the rule it would invoke instead.",
+      "dimension": "application",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/lint-overlay/SKILL.md:102-112"]
+    },
+    {
+      "id": 7,
+      "prompt": "A team asks the lint-overlay to help decide whether their codebase should have lint at all. Should the overlay answer directly? Cite what the overlay says about the scope of decisions it adds vs. the scope it leaves to the base skill.",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "critical",
+      "truth_sources": ["skills/lint-overlay/SKILL.md:65-82", "skills/lint-overlay/SKILL.md:118-124"]
+    },
+    {
+      "id": 8,
+      "prompt": "A CI pipeline currently runs lint across the entire repository tree on every pull request, including files the current change did not touch. According to the lint-overlay's `Lint scope matches the test scope` rule, is this the correct posture? When is tree-wide lint justified, and when should it be restricted to changed files?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/lint-overlay/SKILL.md:102-112", "skills/lint-overlay/SKILL.md:92-100"]
+    },
+    {
+      "id": 9,
+      "prompt": "A team configures their CI so that lint produces output but does not fail the build — the rationale is \"developers will see the warnings and fix them eventually.\" According to the lint-overlay's `Lint failures are blocking, not advisory` rule, is this posture consistent with the overlay's doctrine? What is the overlay's position on warn-only lint, and when (if ever) is it acceptable?",
+      "dimension": "application",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/lint-overlay/SKILL.md:102-112", "skills/lint-overlay/SKILL.md:92-100"]
+    },
+    {
+      "id": 10,
+      "prompt": "A team wants to introduce `react-hooks/exhaustive-deps` as a new ESLint rule. The codebase currently has ~150 files that would newly fail this rule, across 8 feature teams, on code paths that range from new (<1 month old) to legacy (>3 years, unowned). According to the lint-overlay's `New lint rules require a migration plan` rule, produce the migration plan the overlay would require — including phased gates, cleanup ownership, timeline, and the explicit artifact(s) the plan must ship with. Distinguish this from the adjacent `Only enforce rules that were green before the change` rule: what does the migration plan add beyond pinning pre-change green state?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "workflow",
+      "criticality": "high",
+      "truth_sources": ["skills/lint-overlay/SKILL.md:102-112"]
+    }
+  ]
+}

package/examples/evals/microcopy.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "microcopy",
+  "subject": "Functional UI text for action labels, empty states, tooltips, dialogs, loading messages, toasts, validation, permissions, and onboarding steps",
+  "adjacent_concepts": ["semiotics", "a11y", "interaction-feedback", "writing-humanizer"],
+  "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 destructive dialog says Submit and the team needs a specific, consequence-first action label. Which skill owns the wording pattern?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/microcopy/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "An empty state needs to acknowledge what is missing, explain why, and guide the next action in one compact UI surface. What should microcopy produce?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/microcopy/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "The task is marketing persuasion for the pricing-page headline. Should microcopy accept the work?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/microcopy/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "The task is to decide the loading, success, retry, and undo state sequence, not the words inside those states. Which boundary should microcopy respect?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/microcopy/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/observability-modeling.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "observability-modeling",
+  "subject": "Telemetry semantics for diagnostic questions, logs, metrics, traces, spans, attributes, correlation IDs, SLOs, alert signals, cardinality, privacy, and sampling",
+  "adjacent_concepts": ["error-tracking", "event-contract-design", "performance-engineering", "debugging"],
+  "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 background ingestion pipeline needs logs, metrics, spans, correlation IDs, and alert signals designed before implementation. Which skill owns the telemetry model?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/observability-modeling/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "The team has diagnostic questions but no event attributes or SLO signals. What should observability-modeling turn those questions into?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/observability-modeling/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "The task is to configure Sentry error boundaries and sanitized exception capture only. Should observability-modeling accept?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/observability-modeling/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "There is a live production incident and the user wants root-cause debugging. Which boundary should observability-modeling respect?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/observability-modeling/SKILL.md"]
+    }
+  ]
+}

package/examples/evals/pattern-recognition.json

@@ -0,0 +1,96 @@
+{
+  "skill_name": "pattern-recognition",
+  "subject": "Recurring-structure recognition across code, errors, board state, conventions, and domain encodings",
+  "adjacent_concepts": ["debugging", "code-review", "diagnosis", "lint-overlay"],
+  "grounding_note": "This portable skill is grounded in its own method description and cited public pattern-recognition sources rather than repo-local implementation files. These evals exercise the recurring-instance threshold, clustering discipline, prevention loop, lifecycle states, and negative-routing boundaries.",
+  "evals": [
+    {
+      "id": 1,
+      "prompt": "The same SQL string interpolation bug appears in three route handlers. Should the agent fix only the latest one?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/pattern-recognition/SKILL.md:61-80"]
+    },
+    {
+      "id": 2,
+      "prompt": "An agent saw one strange failure once in the logs and wants to write a lint rule immediately. Is that pattern recognition?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/pattern-recognition/SKILL.md:72-80", "skills/pattern-recognition/SKILL.md:462-481"]
+    },
+    {
+      "id": 3,
+      "prompt": "Twelve component files contain hardcoded hex colors outside the design-token source. What is the pattern-recognition response?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/pattern-recognition/SKILL.md:233-279"]
+    },
+    {
+      "id": 4,
+      "prompt": "A session log contains many errors that differ only by timestamps, file paths, and line numbers. How should the agent reason about them?",
+      "dimension": "mental_model",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/pattern-recognition/SKILL.md:155-187"]
+    },
+    {
+      "id": 5,
+      "prompt": "A board has many stale In Progress tasks, WIP overflow, and duplicate-looking tickets. Is that random clutter?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/pattern-recognition/SKILL.md:197-231"]
+    },
+    {
+      "id": 6,
+      "prompt": "The same dashboard NaN display bug keeps coming back. A developer proposes adding `|| 0` at every callsite. What should pattern-recognition say?",
+      "dimension": "purpose",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "critical",
+      "truth_sources": ["skills/pattern-recognition/SKILL.md:281-329", "skills/pattern-recognition/SKILL.md:366-395"]
+    },
+    {
+      "id": 7,
+      "prompt": "A detection rule for an old import path now returns zero matches every week after the migration finished. What state is that pattern in?",
+      "dimension": "taxonomy",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/pattern-recognition/SKILL.md:397-429"]
+    },
+    {
+      "id": 8,
+      "prompt": "The user asks: 'Review this PR for semantic correctness and maintainability.' Should this skill take over?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/pattern-recognition/SKILL.md:483-494"]
+    }
+  ]
+}

package/examples/evals/performance-engineering.json

@@ -0,0 +1,52 @@
+{
+  "skill_name": "performance-engineering",
+  "subject": "Performance measurement and improvement for latency, throughput, Core Web Vitals, database queries, caching, bundle size, concurrency, resource use, and regressions",
+  "adjacent_concepts": ["observability-modeling", "testing-strategy", "api-design", "data-modeling"],
+  "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 dashboard is slow and the team needs measurement, bottleneck analysis, and a decision about what to optimize first. Which skill owns the work?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "process",
+      "truth_mode": "process_correctness",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/performance-engineering/SKILL.md"]
+    },
+    {
+      "id": 2,
+      "prompt": "An endpoint risks N+1 queries and cache mistakes, and the team wants regression prevention. What should performance-engineering verify?",
+      "dimension": "application",
+      "substance": "domain",
+      "calibration": "semantic",
+      "truth_mode": "conceptual_correctness_plus_repo_application",
+      "skill_type": "concept",
+      "criticality": "high",
+      "truth_sources": ["skills/performance-engineering/SKILL.md"]
+    },
+    {
+      "id": 3,
+      "prompt": "The task is to design logs, metrics, spans, and correlation IDs before implementation, not optimize a measured bottleneck. Should performance-engineering accept?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/performance-engineering/SKILL.md"]
+    },
+    {
+      "id": 4,
+      "prompt": "The user asks for random micro-optimizations without a measurement or bottleneck. Which performance-engineering guardrail applies?",
+      "dimension": "boundary",
+      "substance": "contradiction-check",
+      "calibration": "semantic",
+      "truth_mode": "code_verification",
+      "skill_type": "concept",
+      "criticality": "normal",
+      "truth_sources": ["skills/performance-engineering/SKILL.md"]
+    }
+  ]
+}