@dzhechkov/harness-core 0.2.0 → 0.3.1

package/dist/create-skill.d.ts

@@ -0,0 +1,39 @@
+/**
+ * Skill scaffolder — creates a complete SKILL.md directory structure.
+ *
+ * Generates: SKILL.md (agentskills.io frontmatter), schemas/output.json,
+ * scripts/validate-config.json. Optionally: evals/, references/.
+ *
+ * @packageDocumentation
+ */
+/** Options for skill creation. */
+export interface CreateSkillOptions {
+    /** Skill id (kebab-case directory name). */
+    readonly name: string;
+    /** One-line description for SKILL.md frontmatter. */
+    readonly description: string;
+    /** Parent directory where skill dir will be created. Default: `.claude/skills`. */
+    readonly skillsDir?: string | undefined;
+    /** Include evals/ directory with template. */
+    readonly withEvals?: boolean | undefined;
+    /** Include references/ directory. */
+    readonly withReferences?: boolean | undefined;
+    /** Trust tier (1-3). Default: 1. */
+    readonly trustTier?: number | undefined;
+    /** Generate BTO-compatible eval templates with 3-layer benchmarks. */
+    readonly bto?: boolean | undefined;
+}
+/** Result of skill creation. */
+export interface CreateSkillResult {
+    readonly skillDir: string;
+    readonly filesCreated: readonly string[];
+    readonly alreadyExists: boolean;
+}
+/**
+ * Create a new skill directory with all required files.
+ *
+ * Returns the list of created files. If the skill directory already exists,
+ * returns `alreadyExists: true` and creates nothing.
+ */
+export declare function createSkill(opts: CreateSkillOptions): CreateSkillResult;
+//# sourceMappingURL=create-skill.d.ts.map

package/dist/create-skill.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"create-skill.d.ts","sourceRoot":"","sources":["../src/create-skill.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAKH,kCAAkC;AAClC,MAAM,WAAW,kBAAkB;IACjC,4CAA4C;IAC5C,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,qDAAqD;IACrD,QAAQ,CAAC,WAAW,EAAE,MAAM,CAAC;IAC7B,mFAAmF;IACnF,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACxC,8CAA8C;IAC9C,QAAQ,CAAC,SAAS,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IACzC,qCAAqC;IACrC,QAAQ,CAAC,cAAc,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9C,oCAAoC;IACpC,QAAQ,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IACxC,sEAAsE;IACtE,QAAQ,CAAC,GAAG,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;CACpC;AAED,gCAAgC;AAChC,MAAM,WAAW,iBAAiB;IAChC,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B,QAAQ,CAAC,YAAY,EAAE,SAAS,MAAM,EAAE,CAAC;IACzC,QAAQ,CAAC,aAAa,EAAE,OAAO,CAAC;CACjC;AAiQD;;;;;GAKG;AACH,wBAAgB,WAAW,CAAC,IAAI,EAAE,kBAAkB,GAAG,iBAAiB,CA+CvE"}

package/dist/create-skill.js

@@ -0,0 +1,306 @@
+/**
+ * Skill scaffolder — creates a complete SKILL.md directory structure.
+ *
+ * Generates: SKILL.md (agentskills.io frontmatter), schemas/output.json,
+ * scripts/validate-config.json. Optionally: evals/, references/.
+ *
+ * @packageDocumentation
+ */
+import { existsSync, mkdirSync, writeFileSync } from 'node:fs';
+import { join } from 'node:path';
+/** Generate SKILL.md content with agentskills.io frontmatter. */
+function generateSkillMd(opts) {
+    const tier = opts.trustTier ?? 1;
+    return `---
+name: "${opts.name}"
+description: "${opts.description}"
+trust_tier: ${tier}
+trust_tier_label: "${tier === 3 ? 'Verified' : tier === 2 ? 'Validated' : 'Structured'}"
+validation:
+  schema_path: schemas/output.json
+  validator_path: scripts/validate-config.json
+---
+
+# ${opts.name}
+
+${opts.description}
+
+## When to use
+
+Use when [describe the trigger conditions].
+
+## Protocol
+
+1. [Step 1]
+2. [Step 2]
+3. [Step 3]
+
+## Output
+
+Structured output per \`schemas/output.json\`.
+`;
+}
+/** Generate output schema template. */
+function generateOutputSchema(name) {
+    return JSON.stringify({
+        $schema: 'http://json-schema.org/draft-07/schema#',
+        title: `${name} output`,
+        type: 'object',
+        properties: {
+            status: { type: 'string', enum: ['success', 'failure', 'partial'] },
+            summary: { type: 'string' },
+            artifacts: { type: 'array', items: { type: 'string' } },
+        },
+        required: ['status', 'summary'],
+    }, null, 2);
+}
+/** Generate validate-config.json template. */
+function generateValidateConfig(name) {
+    return JSON.stringify({
+        $schema: 'http://json-schema.org/draft-07/schema#',
+        title: `${name} config validation`,
+        type: 'object',
+        properties: {
+            enabled: { type: 'boolean', default: true },
+        },
+    }, null, 2);
+}
+/** Generate eval template. */
+function generateEval(name) {
+    return `# ${name} evaluation
+# Run: aqe eval ${name}
+
+test_cases:
+  - name: "basic invocation"
+    input: "Run ${name}"
+    expected:
+      status: success
+      summary_contains: "${name}"
+`;
+}
+/** Generate BTO-compatible eval template with 3-layer benchmarks. */
+function generateBtoEval(name) {
+    return `# ${name} — BTO evaluation
+# Run: /bto-test .claude/skills/${name}
+# Layers: L0 (deterministic) → L1 (single judge) → L2 (3-judge panel)
+
+artifact_type: skill
+bto_version: "1.0"
+
+# ── Layer 0: Deterministic Pre-checks (free, always runs) ──────────
+layer_0:
+  gate_threshold: 80  # percent pass rate to proceed
+  checks:
+    # Universal checks (U1-U5)
+    - id: U1
+      name: "file exists"
+      check: "SKILL.md exists in skill directory"
+    - id: U2
+      name: "valid UTF-8"
+      check: "SKILL.md is valid UTF-8"
+    - id: U3
+      name: "has headings"
+      check: "SKILL.md contains at least one markdown heading"
+    - id: U4
+      name: "no excessive blanks"
+      check: "no more than 3 consecutive blank lines"
+    - id: U5
+      name: "size bounds"
+      check: "file size between 100 bytes and 50KB"
+    # Skill-specific checks (S1-S10)
+    - id: S1
+      name: "frontmatter present"
+      check: "SKILL.md starts with YAML frontmatter (---)"
+    - id: S2
+      name: "name field"
+      check: "frontmatter contains name field"
+    - id: S3
+      name: "description field"
+      check: "frontmatter contains description field"
+    - id: S4
+      name: "trust_tier field"
+      check: "frontmatter contains trust_tier (1-3)"
+    - id: S5
+      name: "protocol section"
+      check: "SKILL.md contains ## Protocol or ## Steps"
+    - id: S6
+      name: "output section"
+      check: "SKILL.md contains ## Output"
+    - id: S7
+      name: "schema exists"
+      check: "schemas/output.json exists and is valid JSON"
+    - id: S8
+      name: "validator exists"
+      check: "scripts/validate-config.json exists"
+    - id: S9
+      name: "when to use"
+      check: "SKILL.md contains ## When to use"
+    - id: S10
+      name: "no TODO placeholders"
+      check: "no [TODO] or [PLACEHOLDER] markers in SKILL.md"
+
+# ── Layer 1: Single LLM Judge (Haiku, quick) ──────────────────────
+layer_1:
+  model: haiku
+  pass_threshold: 7.0
+  dimensions:
+    - name: CLARITY
+      weight: 1.0
+      anchors:
+        "9-10": "Instructions unambiguous, no interpretation needed"
+        "5-6": "Mostly clear but some steps need clarification"
+        "1-3": "Confusing, contradictory, or missing instructions"
+    - name: COMPLETENESS
+      weight: 1.0
+      anchors:
+        "9-10": "All sections filled, protocol covers edge cases"
+        "5-6": "Main path covered, edge cases missing"
+        "1-3": "Stub-level, most sections empty"
+    - name: ACTIONABILITY
+      weight: 1.0
+      anchors:
+        "9-10": "Each step produces a concrete, verifiable output"
+        "5-6": "Some steps vague or unmeasurable"
+        "1-3": "Steps are descriptions, not actions"
+    - name: QUALITY
+      weight: 1.0
+      anchors:
+        "9-10": "Production-ready, well-structured, follows conventions"
+        "5-6": "Functional but needs polish"
+        "1-3": "Draft quality, significant issues"
+    - name: ANTI_PATTERNS
+      weight: 1.0
+      anchors:
+        "9-10": "No anti-patterns detected"
+        "5-6": "Minor anti-patterns (e.g., wall of text)"
+        "1-3": "Major anti-patterns (e.g., no error handling, no abort)"
+
+# ── Layer 2: Full Judge Panel (3 × Sonnet, deep) ──────────────────
+layer_2:
+  judges:
+    - role: expert
+      weight: 0.40
+      focus: "methodology, depth, correctness, domain fit"
+    - role: critic
+      weight: 0.30
+      focus: "gaps, weaknesses, anti-patterns, failure modes"
+    - role: auditor
+      weight: 0.30
+      focus: "structure, coverage, cross-references, completeness"
+  model: sonnet
+  pass_threshold: 7.0
+  dimensions:
+    - METHODOLOGY
+    - DEPTH
+    - CORRECTNESS
+    - USABILITY
+    - ROBUSTNESS
+  disagreement_threshold: 3  # max-min > 3 triggers meta-judge
+
+# ── Quality Gates ──────────────────────────────────────────────────
+gates:
+  layer_0: "pass_rate >= 80%"
+  layer_1: "average >= 7.0"
+  layer_2: "weighted_average >= 7.0"
+  optimization_delta: 0.5  # min improvement per iteration
+  max_iterations: 10
+
+# ── Test Cases ─────────────────────────────────────────────────────
+test_cases:
+  - name: "basic invocation"
+    input: "Run ${name}"
+    expected:
+      status: success
+      summary_contains: "${name}"
+  - name: "edge case — empty input"
+    input: ""
+    expected:
+      status: failure
+      summary_contains: "input required"
+  - name: "edge case — malformed input"
+    input: "{{RANDOM_GARBAGE}}"
+    expected:
+      status: failure
+      summary_contains: "invalid"
+`;
+}
+/** Generate BTO judge rubrics reference file. */
+function generateBtoRubrics(name) {
+    return `# ${name} — Judge Rubrics
+
+## Evaluation Dimensions (Layer 2)
+
+| Dimension | Expert (0.40) | Critic (0.30) | Auditor (0.30) |
+|-----------|--------------|---------------|----------------|
+| METHODOLOGY | Multi-step protocol with decision points | Missing decision branches | Steps are numbered and sequential |
+| DEPTH | Detailed modules, references, examples | Stub sections, no examples | All sections present and non-empty |
+| CORRECTNESS | Instructions produce expected output | Wrong output or side effects | Schema matches actual output |
+| USABILITY | Quick start, clear navigation | Wall of text, no structure | Cross-references work, paths valid |
+| ROBUSTNESS | Anti-patterns, failure modes, abort | No error handling | Failure modes documented |
+
+## Scoring Anchors
+
+| Score | Label | Description |
+|-------|-------|-------------|
+| 9-10 | Excellent | Production-ready, no changes needed |
+| 7-8 | Good | Minor improvements possible |
+| 5-6 | Needs work | Functional but gaps exist |
+| 3-4 | Poor | Significant rework required |
+| 1-2 | Failed | Does not meet minimum standards |
+
+## Anti-Patterns to Flag
+
+- Score inflation: all judges score >8.5 on first attempt
+- Conformity collapse: identical scores across all judges
+- Missing rejection log: failed checks silently skipped
+- Phantom improvement: score delta >0.5 but no content change
+`;
+}
+/**
+ * Create a new skill directory with all required files.
+ *
+ * Returns the list of created files. If the skill directory already exists,
+ * returns `alreadyExists: true` and creates nothing.
+ */
+export function createSkill(opts) {
+    const skillsDir = opts.skillsDir ?? '.claude/skills';
+    const skillDir = join(skillsDir, opts.name);
+    const filesCreated = [];
+    if (existsSync(skillDir)) {
+        return { skillDir, filesCreated: [], alreadyExists: true };
+    }
+    // Create directories
+    mkdirSync(join(skillDir, 'schemas'), { recursive: true });
+    mkdirSync(join(skillDir, 'scripts'), { recursive: true });
+    // SKILL.md
+    const skillMdPath = join(skillDir, 'SKILL.md');
+    writeFileSync(skillMdPath, generateSkillMd(opts));
+    filesCreated.push('SKILL.md');
+    // schemas/output.json
+    writeFileSync(join(skillDir, 'schemas', 'output.json'), generateOutputSchema(opts.name));
+    filesCreated.push('schemas/output.json');
+    // scripts/validate-config.json
+    writeFileSync(join(skillDir, 'scripts', 'validate-config.json'), generateValidateConfig(opts.name));
+    filesCreated.push('scripts/validate-config.json');
+    // Optional: evals/
+    if (opts.withEvals !== false) {
+        mkdirSync(join(skillDir, 'evals'), { recursive: true });
+        const evalContent = opts.bto ? generateBtoEval(opts.name) : generateEval(opts.name);
+        writeFileSync(join(skillDir, 'evals', `${opts.name}.yaml`), evalContent);
+        filesCreated.push(`evals/${opts.name}.yaml`);
+    }
+    // Optional: references/
+    if (opts.withReferences || opts.bto) {
+        mkdirSync(join(skillDir, 'references'), { recursive: true });
+        if (opts.bto) {
+            writeFileSync(join(skillDir, 'references', 'judge-rubrics.md'), generateBtoRubrics(opts.name));
+            filesCreated.push('references/judge-rubrics.md');
+        }
+        else {
+            writeFileSync(join(skillDir, 'references', '.gitkeep'), '');
+            filesCreated.push('references/.gitkeep');
+        }
+    }
+    return { skillDir, filesCreated, alreadyExists: false };
+}
+//# sourceMappingURL=create-skill.js.map

package/dist/create-skill.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"create-skill.js","sourceRoot":"","sources":["../src/create-skill.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAC/D,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AA2BjC,iEAAiE;AACjE,SAAS,eAAe,CAAC,IAAwB;IAC/C,MAAM,IAAI,GAAG,IAAI,CAAC,SAAS,IAAI,CAAC,CAAC;IACjC,OAAO;SACA,IAAI,CAAC,IAAI;gBACF,IAAI,CAAC,WAAW;cAClB,IAAI;qBACG,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,YAAY;;;;;;IAMlF,IAAI,CAAC,IAAI;;EAEX,IAAI,CAAC,WAAW;;;;;;;;;;;;;;;CAejB,CAAC;AACF,CAAC;AAED,uCAAuC;AACvC,SAAS,oBAAoB,CAAC,IAAY;IACxC,OAAO,IAAI,CAAC,SAAS,CAAC;QACpB,OAAO,EAAE,yCAAyC;QAClD,KAAK,EAAE,GAAG,IAAI,SAAS;QACvB,IAAI,EAAE,QAAQ;QACd,UAAU,EAAE;YACV,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,SAAS,EAAE,SAAS,EAAE,SAAS,CAAC,EAAE;YACnE,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;YAC3B,SAAS,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE;SACxD;QACD,QAAQ,EAAE,CAAC,QAAQ,EAAE,SAAS,CAAC;KAChC,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;AACd,CAAC;AAED,8CAA8C;AAC9C,SAAS,sBAAsB,CAAC,IAAY;IAC1C,OAAO,IAAI,CAAC,SAAS,CAAC;QACpB,OAAO,EAAE,yCAAyC;QAClD,KAAK,EAAE,GAAG,IAAI,oBAAoB;QAClC,IAAI,EAAE,QAAQ;QACd,UAAU,EAAE;YACV,OAAO,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,IAAI,EAAE;SAC5C;KACF,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;AACd,CAAC;AAED,8BAA8B;AAC9B,SAAS,YAAY,CAAC,IAAY;IAChC,OAAO,KAAK,IAAI;kBACA,IAAI;;;;kBAIJ,IAAI;;;2BAGK,IAAI;CAC9B,CAAC;AACF,CAAC;AAED,qEAAqE;AACrE,SAAS,eAAe,CAAC,IAAY;IACnC,OAAO,KAAK,IAAI;kCACgB,IAAI;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;kBA+HpB,IAAI;;;2BAGK,IAAI;;;;;;;;;;;CAW9B,CAAC;AACF,CAAC;AAED,iDAAiD;AACjD,SAAS,kBAAkB,CAAC,IAAY;IACtC,OAAO,KAAK,IAAI;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA4BjB,CAAC;AACF,CAAC;AAGD;;;;;GAKG;AACH,MAAM,UAAU,WAAW,CAAC,IAAwB;IAClD,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,gBAAgB,CAAC;IACrD,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,EAAE,IAAI,CAAC,IAAI,CAAC,CAAC;IAC5C,MAAM,YAAY,GAAa,EAAE,CAAC;IAElC,IAAI,UAAU,CAAC,QAAQ,CAAC,EAAE,CAAC;QACzB,OAAO,EAAE,QAAQ,EAAE,YAAY,EAAE,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC;IAC7D,CAAC;IAED,qBAAqB;IACrB,SAAS,CAAC,IAAI,CAAC,QAAQ,EAAE,SAAS,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC1D,SAAS,CAAC,IAAI,CAAC,QAAQ,EAAE,SAAS,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAE1D,WAAW;IACX,MAAM,WAAW,GAAG,IAAI,CAAC,QAAQ,EAAE,UAAU,CAAC,CAAC;IAC/C,aAAa,CAAC,WAAW,EAAE,eAAe,CAAC,IAAI,CAAC,CAAC,CAAC;IAClD,YAAY,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;IAE9B,sBAAsB;IACtB,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,SAAS,EAAE,aAAa,CAAC,EAAE,oBAAoB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IACzF,YAAY,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;IAEzC,+BAA+B;IAC/B,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,SAAS,EAAE,sBAAsB,CAAC,EAAE,sBAAsB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;IACpG,YAAY,CAAC,IAAI,CAAC,8BAA8B,CAAC,CAAC;IAElD,mBAAmB;IACnB,IAAI,IAAI,CAAC,SAAS,KAAK,KAAK,EAAE,CAAC;QAC7B,SAAS,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACxD,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QACpF,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,EAAE,GAAG,IAAI,CAAC,IAAI,OAAO,CAAC,EAAE,WAAW,CAAC,CAAC;QACzE,YAAY,CAAC,IAAI,CAAC,SAAS,IAAI,CAAC,IAAI,OAAO,CAAC,CAAC;IAC/C,CAAC;IAED,wBAAwB;IACxB,IAAI,IAAI,CAAC,cAAc,IAAI,IAAI,CAAC,GAAG,EAAE,CAAC;QACpC,SAAS,CAAC,IAAI,CAAC,QAAQ,EAAE,YAAY,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QAC7D,IAAI,IAAI,CAAC,GAAG,EAAE,CAAC;YACb,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,YAAY,EAAE,kBAAkB,CAAC,EAAE,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC;YAC/F,YAAY,CAAC,IAAI,CAAC,6BAA6B,CAAC,CAAC;QACnD,CAAC;aAAM,CAAC;YACN,aAAa,CAAC,IAAI,CAAC,QAAQ,EAAE,YAAY,EAAE,UAAU,CAAC,EAAE,EAAE,CAAC,CAAC;YAC5D,YAAY,CAAC,IAAI,CAAC,qBAAqB,CAAC,CAAC;QAC3C,CAAC;IACH,CAAC;IAED,OAAO,EAAE,QAAQ,EAAE,YAAY,EAAE,aAAa,EAAE,KAAK,EAAE,CAAC;AAC1D,CAAC"}

package/dist/index.d.ts

@@ -10,4 +10,6 @@ export * from './apply.js';
 export * from './targets.js';
 export * from './operations.js';
 export * from './workflows.js';
+export { createSkill } from './create-skill.js';
+export type { CreateSkillOptions, CreateSkillResult } from './create-skill.js';
 //# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,yDAAyD;AACzD,eAAO,MAAM,oBAAoB,UAAU,CAAC;AAE5C,cAAc,aAAa,CAAC;AAC5B,cAAc,YAAY,CAAC;AAC3B,cAAc,cAAc,CAAC;AAC7B,cAAc,iBAAiB,CAAC;AAChC,cAAc,gBAAgB,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,yDAAyD;AACzD,eAAO,MAAM,oBAAoB,UAAU,CAAC;AAE5C,cAAc,aAAa,CAAC;AAC5B,cAAc,YAAY,CAAC;AAC3B,cAAc,cAAc,CAAC;AAC7B,cAAc,iBAAiB,CAAC;AAChC,cAAc,gBAAgB,CAAC;AAC/B,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAChD,YAAY,EAAE,kBAAkB,EAAE,iBAAiB,EAAE,MAAM,mBAAmB,CAAC"}

package/dist/index.js

@@ -10,4 +10,5 @@ export * from './apply.js';
 export * from './targets.js';
 export * from './operations.js';
 export * from './workflows.js';
+export { createSkill } from './create-skill.js';
 //# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -1 +1 @@
-{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,yDAAyD;AACzD,MAAM,CAAC,MAAM,oBAAoB,GAAG,OAAO,CAAC;AAE5C,cAAc,aAAa,CAAC;AAC5B,cAAc,YAAY,CAAC;AAC3B,cAAc,cAAc,CAAC;AAC7B,cAAc,iBAAiB,CAAC;AAChC,cAAc,gBAAgB,CAAC"}
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,yDAAyD;AACzD,MAAM,CAAC,MAAM,oBAAoB,GAAG,OAAO,CAAC;AAE5C,cAAc,aAAa,CAAC;AAC5B,cAAc,YAAY,CAAC;AAC3B,cAAc,cAAc,CAAC;AAC7B,cAAc,iBAAiB,CAAC;AAChC,cAAc,gBAAgB,CAAC;AAC/B,OAAO,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@dzhechkov/harness-core",
-  "version": "0.2.0",
+  "version": "0.3.1",
   "description": "Shared harness logic - skill loading, additive apply, and the init/sync/verify/doctor operations.",
   "type": "module",
   "license": "MIT",
@@ -25,11 +25,11 @@
   ],
   "dependencies": {
     "yaml": "^2.0.0",
-    "@dzhechkov/adapter-claude": "0.2.0",
     "@dzhechkov/adapter-codex": "0.2.0",
-    "@dzhechkov/adapter-hermes": "0.2.0",
     "@dzhechkov/adapter-opencode": "0.2.0",
-    "@dzhechkov/core": "0.2.0"
+    "@dzhechkov/core": "0.2.0",
+    "@dzhechkov/adapter-claude": "0.2.0",
+    "@dzhechkov/adapter-hermes": "0.2.0"
   },
   "devDependencies": {
     "@types/node": "^25.6.0",

package/src/create-skill.ts

@@ -0,0 +1,346 @@
+/**
+ * Skill scaffolder — creates a complete SKILL.md directory structure.
+ *
+ * Generates: SKILL.md (agentskills.io frontmatter), schemas/output.json,
+ * scripts/validate-config.json. Optionally: evals/, references/.
+ *
+ * @packageDocumentation
+ */
+
+import { existsSync, mkdirSync, writeFileSync } from 'node:fs';
+import { join } from 'node:path';
+
+/** Options for skill creation. */
+export interface CreateSkillOptions {
+  /** Skill id (kebab-case directory name). */
+  readonly name: string;
+  /** One-line description for SKILL.md frontmatter. */
+  readonly description: string;
+  /** Parent directory where skill dir will be created. Default: `.claude/skills`. */
+  readonly skillsDir?: string | undefined;
+  /** Include evals/ directory with template. */
+  readonly withEvals?: boolean | undefined;
+  /** Include references/ directory. */
+  readonly withReferences?: boolean | undefined;
+  /** Trust tier (1-3). Default: 1. */
+  readonly trustTier?: number | undefined;
+  /** Generate BTO-compatible eval templates with 3-layer benchmarks. */
+  readonly bto?: boolean | undefined;
+}
+
+/** Result of skill creation. */
+export interface CreateSkillResult {
+  readonly skillDir: string;
+  readonly filesCreated: readonly string[];
+  readonly alreadyExists: boolean;
+}
+
+/** Generate SKILL.md content with agentskills.io frontmatter. */
+function generateSkillMd(opts: CreateSkillOptions): string {
+  const tier = opts.trustTier ?? 1;
+  return `---
+name: "${opts.name}"
+description: "${opts.description}"
+trust_tier: ${tier}
+trust_tier_label: "${tier === 3 ? 'Verified' : tier === 2 ? 'Validated' : 'Structured'}"
+validation:
+  schema_path: schemas/output.json
+  validator_path: scripts/validate-config.json
+---
+
+# ${opts.name}
+
+${opts.description}
+
+## When to use
+
+Use when [describe the trigger conditions].
+
+## Protocol
+
+1. [Step 1]
+2. [Step 2]
+3. [Step 3]
+
+## Output
+
+Structured output per \`schemas/output.json\`.
+`;
+}
+
+/** Generate output schema template. */
+function generateOutputSchema(name: string): string {
+  return JSON.stringify({
+    $schema: 'http://json-schema.org/draft-07/schema#',
+    title: `${name} output`,
+    type: 'object',
+    properties: {
+      status: { type: 'string', enum: ['success', 'failure', 'partial'] },
+      summary: { type: 'string' },
+      artifacts: { type: 'array', items: { type: 'string' } },
+    },
+    required: ['status', 'summary'],
+  }, null, 2);
+}
+
+/** Generate validate-config.json template. */
+function generateValidateConfig(name: string): string {
+  return JSON.stringify({
+    $schema: 'http://json-schema.org/draft-07/schema#',
+    title: `${name} config validation`,
+    type: 'object',
+    properties: {
+      enabled: { type: 'boolean', default: true },
+    },
+  }, null, 2);
+}
+
+/** Generate eval template. */
+function generateEval(name: string): string {
+  return `# ${name} evaluation
+# Run: aqe eval ${name}
+
+test_cases:
+  - name: "basic invocation"
+    input: "Run ${name}"
+    expected:
+      status: success
+      summary_contains: "${name}"
+`;
+}
+
+/** Generate BTO-compatible eval template with 3-layer benchmarks. */
+function generateBtoEval(name: string): string {
+  return `# ${name} — BTO evaluation
+# Run: /bto-test .claude/skills/${name}
+# Layers: L0 (deterministic) → L1 (single judge) → L2 (3-judge panel)
+
+artifact_type: skill
+bto_version: "1.0"
+
+# ── Layer 0: Deterministic Pre-checks (free, always runs) ──────────
+layer_0:
+  gate_threshold: 80  # percent pass rate to proceed
+  checks:
+    # Universal checks (U1-U5)
+    - id: U1
+      name: "file exists"
+      check: "SKILL.md exists in skill directory"
+    - id: U2
+      name: "valid UTF-8"
+      check: "SKILL.md is valid UTF-8"
+    - id: U3
+      name: "has headings"
+      check: "SKILL.md contains at least one markdown heading"
+    - id: U4
+      name: "no excessive blanks"
+      check: "no more than 3 consecutive blank lines"
+    - id: U5
+      name: "size bounds"
+      check: "file size between 100 bytes and 50KB"
+    # Skill-specific checks (S1-S10)
+    - id: S1
+      name: "frontmatter present"
+      check: "SKILL.md starts with YAML frontmatter (---)"
+    - id: S2
+      name: "name field"
+      check: "frontmatter contains name field"
+    - id: S3
+      name: "description field"
+      check: "frontmatter contains description field"
+    - id: S4
+      name: "trust_tier field"
+      check: "frontmatter contains trust_tier (1-3)"
+    - id: S5
+      name: "protocol section"
+      check: "SKILL.md contains ## Protocol or ## Steps"
+    - id: S6
+      name: "output section"
+      check: "SKILL.md contains ## Output"
+    - id: S7
+      name: "schema exists"
+      check: "schemas/output.json exists and is valid JSON"
+    - id: S8
+      name: "validator exists"
+      check: "scripts/validate-config.json exists"
+    - id: S9
+      name: "when to use"
+      check: "SKILL.md contains ## When to use"
+    - id: S10
+      name: "no TODO placeholders"
+      check: "no [TODO] or [PLACEHOLDER] markers in SKILL.md"
+
+# ── Layer 1: Single LLM Judge (Haiku, quick) ──────────────────────
+layer_1:
+  model: haiku
+  pass_threshold: 7.0
+  dimensions:
+    - name: CLARITY
+      weight: 1.0
+      anchors:
+        "9-10": "Instructions unambiguous, no interpretation needed"
+        "5-6": "Mostly clear but some steps need clarification"
+        "1-3": "Confusing, contradictory, or missing instructions"
+    - name: COMPLETENESS
+      weight: 1.0
+      anchors:
+        "9-10": "All sections filled, protocol covers edge cases"
+        "5-6": "Main path covered, edge cases missing"
+        "1-3": "Stub-level, most sections empty"
+    - name: ACTIONABILITY
+      weight: 1.0
+      anchors:
+        "9-10": "Each step produces a concrete, verifiable output"
+        "5-6": "Some steps vague or unmeasurable"
+        "1-3": "Steps are descriptions, not actions"
+    - name: QUALITY
+      weight: 1.0
+      anchors:
+        "9-10": "Production-ready, well-structured, follows conventions"
+        "5-6": "Functional but needs polish"
+        "1-3": "Draft quality, significant issues"
+    - name: ANTI_PATTERNS
+      weight: 1.0
+      anchors:
+        "9-10": "No anti-patterns detected"
+        "5-6": "Minor anti-patterns (e.g., wall of text)"
+        "1-3": "Major anti-patterns (e.g., no error handling, no abort)"
+
+# ── Layer 2: Full Judge Panel (3 × Sonnet, deep) ──────────────────
+layer_2:
+  judges:
+    - role: expert
+      weight: 0.40
+      focus: "methodology, depth, correctness, domain fit"
+    - role: critic
+      weight: 0.30
+      focus: "gaps, weaknesses, anti-patterns, failure modes"
+    - role: auditor
+      weight: 0.30
+      focus: "structure, coverage, cross-references, completeness"
+  model: sonnet
+  pass_threshold: 7.0
+  dimensions:
+    - METHODOLOGY
+    - DEPTH
+    - CORRECTNESS
+    - USABILITY
+    - ROBUSTNESS
+  disagreement_threshold: 3  # max-min > 3 triggers meta-judge
+
+# ── Quality Gates ──────────────────────────────────────────────────
+gates:
+  layer_0: "pass_rate >= 80%"
+  layer_1: "average >= 7.0"
+  layer_2: "weighted_average >= 7.0"
+  optimization_delta: 0.5  # min improvement per iteration
+  max_iterations: 10
+
+# ── Test Cases ─────────────────────────────────────────────────────
+test_cases:
+  - name: "basic invocation"
+    input: "Run ${name}"
+    expected:
+      status: success
+      summary_contains: "${name}"
+  - name: "edge case — empty input"
+    input: ""
+    expected:
+      status: failure
+      summary_contains: "input required"
+  - name: "edge case — malformed input"
+    input: "{{RANDOM_GARBAGE}}"
+    expected:
+      status: failure
+      summary_contains: "invalid"
+`;
+}
+
+/** Generate BTO judge rubrics reference file. */
+function generateBtoRubrics(name: string): string {
+  return `# ${name} — Judge Rubrics
+
+## Evaluation Dimensions (Layer 2)
+
+| Dimension | Expert (0.40) | Critic (0.30) | Auditor (0.30) |
+|-----------|--------------|---------------|----------------|
+| METHODOLOGY | Multi-step protocol with decision points | Missing decision branches | Steps are numbered and sequential |
+| DEPTH | Detailed modules, references, examples | Stub sections, no examples | All sections present and non-empty |
+| CORRECTNESS | Instructions produce expected output | Wrong output or side effects | Schema matches actual output |
+| USABILITY | Quick start, clear navigation | Wall of text, no structure | Cross-references work, paths valid |
+| ROBUSTNESS | Anti-patterns, failure modes, abort | No error handling | Failure modes documented |
+
+## Scoring Anchors
+
+| Score | Label | Description |
+|-------|-------|-------------|
+| 9-10 | Excellent | Production-ready, no changes needed |
+| 7-8 | Good | Minor improvements possible |
+| 5-6 | Needs work | Functional but gaps exist |
+| 3-4 | Poor | Significant rework required |
+| 1-2 | Failed | Does not meet minimum standards |
+
+## Anti-Patterns to Flag
+
+- Score inflation: all judges score >8.5 on first attempt
+- Conformity collapse: identical scores across all judges
+- Missing rejection log: failed checks silently skipped
+- Phantom improvement: score delta >0.5 but no content change
+`;
+}
+
+
+/**
+ * Create a new skill directory with all required files.
+ *
+ * Returns the list of created files. If the skill directory already exists,
+ * returns `alreadyExists: true` and creates nothing.
+ */
+export function createSkill(opts: CreateSkillOptions): CreateSkillResult {
+  const skillsDir = opts.skillsDir ?? '.claude/skills';
+  const skillDir = join(skillsDir, opts.name);
+  const filesCreated: string[] = [];
+
+  if (existsSync(skillDir)) {
+    return { skillDir, filesCreated: [], alreadyExists: true };
+  }
+
+  // Create directories
+  mkdirSync(join(skillDir, 'schemas'), { recursive: true });
+  mkdirSync(join(skillDir, 'scripts'), { recursive: true });
+
+  // SKILL.md
+  const skillMdPath = join(skillDir, 'SKILL.md');
+  writeFileSync(skillMdPath, generateSkillMd(opts));
+  filesCreated.push('SKILL.md');
+
+  // schemas/output.json
+  writeFileSync(join(skillDir, 'schemas', 'output.json'), generateOutputSchema(opts.name));
+  filesCreated.push('schemas/output.json');
+
+  // scripts/validate-config.json
+  writeFileSync(join(skillDir, 'scripts', 'validate-config.json'), generateValidateConfig(opts.name));
+  filesCreated.push('scripts/validate-config.json');
+
+  // Optional: evals/
+  if (opts.withEvals !== false) {
+    mkdirSync(join(skillDir, 'evals'), { recursive: true });
+    const evalContent = opts.bto ? generateBtoEval(opts.name) : generateEval(opts.name);
+    writeFileSync(join(skillDir, 'evals', `${opts.name}.yaml`), evalContent);
+    filesCreated.push(`evals/${opts.name}.yaml`);
+  }
+
+  // Optional: references/
+  if (opts.withReferences || opts.bto) {
+    mkdirSync(join(skillDir, 'references'), { recursive: true });
+    if (opts.bto) {
+      writeFileSync(join(skillDir, 'references', 'judge-rubrics.md'), generateBtoRubrics(opts.name));
+      filesCreated.push('references/judge-rubrics.md');
+    } else {
+      writeFileSync(join(skillDir, 'references', '.gitkeep'), '');
+      filesCreated.push('references/.gitkeep');
+    }
+  }
+
+  return { skillDir, filesCreated, alreadyExists: false };
+}

package/src/index.ts

@@ -12,3 +12,5 @@ export * from './apply.js';
 export * from './targets.js';
 export * from './operations.js';
 export * from './workflows.js';
+export { createSkill } from './create-skill.js';
+export type { CreateSkillOptions, CreateSkillResult } from './create-skill.js';