@skilljack/evals 1.0.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Ola Hungerford
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,327 @@
+# skilljack-evals
+
+CLI for evaluating AI agent skills. Tests how well agents discover, load, and execute [Agent Skills](https://agentskills.io/home) — measuring discoverability, instruction adherence, and output quality.
+
+Runs standalone or as a GitHub Action.
+
+## What are Agent Skills?
+
+Agent Skills are a lightweight, open-source format for extending AI agent capabilities. Each skill is a folder containing a `SKILL.md` file with metadata and instructions that agents can discover and use. Learn more at [agentskills.io](https://agentskills.io/home).
+
+## Requirements
+
+- Node.js >= 20.0.0
+- Anthropic API key (or AWS credentials for Bedrock)
+
+## Installation
+
+```bash
+npm install
+npm run build
+```
+
+## Quick Start
+
+```bash
+# Run the example greeting evaluation
+skilljack-evals run evals/example-greeting/tasks.yaml --verbose
+
+# Deterministic scoring only (no LLM judge, free)
+skilljack-evals run evals/example-greeting/tasks.yaml --no-judge
+
+# Validate a task file without running
+skilljack-evals validate evals/example-greeting/tasks.yaml
+```
+
+## Configuration
+
+### API Key
+
+Set `ANTHROPIC_API_KEY` in your environment or a `.env` file (see `.env.example`).
+
+### Bedrock
+
+Set these environment variables — the Agent SDK handles the rest:
+
+```bash
+CLAUDE_CODE_USE_BEDROCK=1
+AWS_REGION=us-west-2
+AWS_PROFILE=your-profile
+```
+
+### Config File
+
+Create an `eval.config.yaml` in your project root (all fields optional):
+
+```yaml
+models:
+  agent: sonnet        # EVAL_AGENT_MODEL
+  judge: haiku         # EVAL_JUDGE_MODEL
+
+scoring:
+  weights:
+    discovery: 0.3
+    adherence: 0.4
+    output: 0.3
+
+thresholds:
+  discovery_rate: 0.8  # EVAL_DISCOVERY_THRESHOLD
+  avg_score: 4.0       # EVAL_SCORE_THRESHOLD
+
+runner:
+  timeout_ms: 300000   # EVAL_TASK_TIMEOUT_MS
+  allowed_write_dirs:
+    - ./results/
+    - ./fixtures/
+
+output:
+  dir: ./results       # EVAL_OUTPUT_DIR
+  judge_truncation: 5000
+  report_truncation: 2000
+
+ci:
+  exit_on_failure: true
+  github_summary: false
+```
+
+**Precedence** (lowest to highest): YAML defaults → `eval.config.yaml` → environment variables (`EVAL_*`) → CLI flags.
+
+## CLI Commands
+
+### `run` — Full evaluation pipeline
+
+Runs the agent against tasks, scores results, and generates reports.
+
+```bash
+skilljack-evals run evals/greeting/tasks.yaml \
+  --model sonnet --judge-model haiku \
+  --timeout 300000 \
+  --tasks gr-001,gr-002 \
+  --threshold-discovery 0.8 --threshold-score 4.0 \
+  --output-dir ./results \
+  --github-summary --verbose
+```
+
+### `score` — Score existing results
+
+```bash
+skilljack-evals score results.json --judge-model haiku
+```
+
+### `report` — Generate reports from scored results
+
+```bash
+skilljack-evals report results.json -o report.md --json report.json
+```
+
+### `validate` — Check YAML syntax
+
+```bash
+skilljack-evals validate evals/greeting/tasks.yaml
+```
+
+### `create-eval` — Generate task template
+
+```bash
+skilljack-evals create-eval greeting -o evals/greeting/tasks.yaml -n 10
+```
+
+### `parse` — Parse YAML to JSON
+
+```bash
+skilljack-evals parse evals/greeting/tasks.yaml
+```
+
+## Architecture
+
+```
+YAML tasks → Config → Runner (Agent SDK) → Scorer (deterministic + LLM judge) → Report
+```
+
+### Pipeline
+
+1. **Parse** — Load and validate task definitions from YAML
+2. **Setup** — Copy skills to `.claude/skills/` in the working directory
+3. **Run** — Execute agent against each task via the Claude Agent SDK
+4. **Score** — Deterministic checks (free, fast) then optional LLM judge
+5. **Report** — Generate markdown + JSON reports, check pass/fail thresholds
+6. **Cleanup** — Remove copied skills
+
+### Scoring
+
+Two scoring methods that can run independently or together:
+
+**Deterministic** (free, fast):
+- Checks tool calls for skill activation
+- Searches output for expected marker strings
+- Validates expected/forbidden tool usage
+- Binary pass/fail
+
+**LLM Judge** (richer, ~$0.001/task):
+- Discovery (0 or 1) — Did the agent load the expected skill?
+- Adherence (1-5) — How well did the agent follow skill instructions?
+- Output Quality (1-5) — Does the output meet task requirements?
+- Failure categorization
+
+**Combined score**: `w_d * discovery + w_a * ((adherence-1)/4) + w_o * ((outputQuality-1)/4)`
+
+### Failure Categories
+
+| Category | Meaning |
+|----------|---------|
+| `discovery_failure` | Agent didn't load the skill |
+| `false_positive` | Agent loaded a skill it shouldn't have |
+| `instruction_ambiguity` | Agent misinterpreted instructions |
+| `missing_guidance` | Skill didn't cover the needed case |
+| `agent_error` | Agent made a mistake despite guidance |
+| `none` | No failure |
+
+## Task File Format
+
+```yaml
+skill: greeting
+version: "1.0"
+
+defaults:
+  expected_skill_load: greeting
+  criteria:
+    discovery: { weight: 0.3 }
+    adherence: { weight: 0.4 }
+    output: { weight: 0.3 }
+
+tasks:
+  - id: gr-001
+    prompt: "Hello! Please greet me using the greeting skill."
+
+    # Deterministic checks (optional, free)
+    deterministic:
+      expect_skill_activation: true
+      expect_marker: "GREETING_SUCCESS"
+      expect_tool_calls: []
+      expect_no_tool_calls: []
+
+    # LLM judge criteria (optional, costs API calls)
+    criteria:
+      discovery: { weight: 0.3, description: "Should load greeting skill" }
+      adherence: { weight: 0.4, description: "Should follow skill format" }
+      output: { weight: 0.3, description: "Greeting is friendly" }
+    golden_checklist:
+      - "Loaded the greeting skill"
+      - "Friendly tone"
+
+  # False positive test — skill should NOT activate
+  - id: gr-fp-001
+    prompt: "What are best practices for email greetings?"
+    expected_skill_load: none
+    deterministic:
+      expect_skill_activation: false
+```
+
+Both `deterministic` and `criteria` blocks are optional. If both are present, the scorer runs both and merges results.
+
+## GitHub Action
+
+```yaml
+- uses: olaservo/skilljack-evals@v1
+  with:
+    tasks: evals/commit/tasks.yaml
+    threshold-discovery: '0.8'
+    threshold-score: '4.0'
+  env:
+    ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
+```
+
+### Inputs
+
+| Input | Required | Default | Description |
+|-------|----------|---------|-------------|
+| `tasks` | Yes | — | Path to tasks YAML file |
+| `model` | No | `sonnet` | Agent model |
+| `judge-model` | No | `haiku` | LLM judge model |
+| `config` | No | — | Path to eval.config.yaml |
+| `threshold-discovery` | No | `0.8` | Minimum discovery rate (0-1) |
+| `threshold-score` | No | `4.0` | Minimum average score (1-5) |
+| `timeout` | No | `300000` | Per-task timeout (ms) |
+| `tasks-filter` | No | — | Comma-separated task IDs |
+| `skills-dir` | No | — | Path to skills directory |
+| `no-judge` | No | `false` | Skip LLM judge |
+| `no-deterministic` | No | `false` | Skip deterministic scoring |
+
+### Outputs
+
+| Output | Description |
+|--------|-------------|
+| `passed` | Whether all thresholds were met |
+| `discovery-rate` | Discovery rate achieved (0-1) |
+| `avg-score` | Average weighted score |
+| `report-path` | Path to markdown report |
+| `json-path` | Path to JSON report |
+
+The action writes a condensed summary to `$GITHUB_STEP_SUMMARY` and exits with code 1 if thresholds are not met.
+
+## Library Usage
+
+```typescript
+import {
+  parseSkillEvaluation,
+  SkillJudge,
+  generateReport,
+  runPipeline,
+  scoreDeterministic,
+  loadConfig,
+} from '@skilljack/evals';
+
+// Full pipeline
+const result = await runPipeline('evals/greeting/tasks.yaml', {
+  model: 'sonnet',
+  verbose: true,
+});
+
+// Or individual steps
+const evaluation = await parseSkillEvaluation('path/to/tasks.yaml');
+const judge = new SkillJudge({ model: 'haiku' });
+const score = await judge.judgeResult(task, result);
+const detScore = scoreDeterministic(task, result);
+const report = generateReport(evaluation, results, scores);
+```
+
+## Development
+
+```bash
+npm run dev        # Run CLI in dev mode (tsx)
+npm run build      # Compile TypeScript
+npm run typecheck  # Type check without emitting
+npm run start      # Run compiled CLI
+```
+
+## Project Structure
+
+```
+src/
+  cli.ts                    # CLI entry point (commander)
+  index.ts                  # Public API exports
+  types.ts                  # TypeScript interfaces
+  config.ts                 # Centralized config (file + env + CLI)
+  parser.ts                 # YAML parsing and validation
+  pipeline.ts               # Full eval pipeline orchestrator
+  runner/
+    runner.ts               # Agent SDK runner
+    skill-setup.ts          # Skill file management
+    security.ts             # Tool write restrictions
+  scorer/
+    scorer.ts               # Score orchestrator (deterministic + judge)
+    deterministic.ts        # Marker/tool-call checks
+    judge.ts                # LLM-as-judge scoring
+  session/
+    session-logger.ts       # Event capture and logging
+  report/
+    report.ts               # Markdown + JSON report generation
+    github-summary.ts       # GitHub Actions summary
+action/
+  action.yml                # GitHub Action metadata
+  index.ts                  # Action entry point
+evals/
+  example-greeting/         # Example evaluation
+    tasks.yaml
+    skills/greeting/SKILL.md
+```

package/action/action.yml

@@ -0,0 +1,72 @@
+name: 'Agent Skill Evaluation'
+description: 'Run agent skill evaluations to test discoverability, adherence, and output quality'
+branding:
+  icon: 'check-circle'
+  color: 'blue'
+
+inputs:
+  tasks:
+    description: 'Path to tasks YAML file'
+    required: true
+  model:
+    description: 'Agent model for task execution'
+    required: false
+    default: 'sonnet'
+  judge-model:
+    description: 'Model for LLM-as-judge scoring'
+    required: false
+    default: 'haiku'
+  config:
+    description: 'Path to eval.config.yaml'
+    required: false
+  threshold-discovery:
+    description: 'Minimum discovery rate (0-1)'
+    required: false
+    default: '0.8'
+  threshold-score:
+    description: 'Minimum average score (1-5)'
+    required: false
+    default: '4.0'
+  timeout:
+    description: 'Per-task timeout in milliseconds'
+    required: false
+    default: '300000'
+  tasks-filter:
+    description: 'Comma-separated list of task IDs to run'
+    required: false
+  skills-dir:
+    description: 'Path to skills directory for local setup'
+    required: false
+  working-directory:
+    description: 'Working directory for agent execution'
+    required: false
+    default: '.'
+  no-judge:
+    description: 'Skip LLM judge scoring (deterministic only)'
+    required: false
+    default: 'false'
+  no-deterministic:
+    description: 'Skip deterministic scoring (LLM judge only)'
+    required: false
+    default: 'false'
+  anthropic-api-key:
+    description: 'Anthropic API key (or use ANTHROPIC_API_KEY env var)'
+    required: false
+
+outputs:
+  passed:
+    description: 'Whether all thresholds were met (true/false)'
+  discovery-rate:
+    description: 'Discovery rate achieved (0-1)'
+  avg-score:
+    description: 'Average weighted score achieved'
+  report-path:
+    description: 'Path to generated markdown report'
+  json-path:
+    description: 'Path to generated JSON report'
+
+runs:
+  using: 'node20'
+  main: '../dist/action/index.js'
+  # Note: For GitHub Actions marketplace, this path should be relative to repo root.
+  # If publishing as a GitHub Action, the main path is: dist/action/index.js

package/action/index.ts

@@ -0,0 +1,78 @@
+/**
+ * GitHub Action entry point for skill evaluation.
+ *
+ * Reads inputs from the action.yml, runs the evaluation pipeline,
+ * and sets outputs + job summary.
+ */
+
+import * as core from '@actions/core';
+import { runPipeline } from '../src/pipeline.js';
+import type { EvalConfig } from '../src/config.js';
+
+async function run(): Promise<void> {
+  try {
+    // Read inputs
+    const tasks = core.getInput('tasks', { required: true });
+    const model = core.getInput('model') || 'sonnet';
+    const judgeModel = core.getInput('judge-model') || 'haiku';
+    const configPath = core.getInput('config') || undefined;
+    const thresholdDiscovery = parseFloat(core.getInput('threshold-discovery') || '0.8');
+    const thresholdScore = parseFloat(core.getInput('threshold-score') || '4.0');
+    const timeout = parseInt(core.getInput('timeout') || '300000', 10);
+    const tasksFilter = core.getInput('tasks-filter') || undefined;
+    const skillsDir = core.getInput('skills-dir') || undefined;
+    const cwd = core.getInput('working-directory') || process.cwd();
+    const noJudge = core.getInput('no-judge') === 'true';
+    const noDeterministic = core.getInput('no-deterministic') === 'true';
+
+    // Handle API key
+    const apiKey = core.getInput('anthropic-api-key') || process.env.ANTHROPIC_API_KEY;
+    if (apiKey) {
+      process.env.ANTHROPIC_API_KEY = apiKey;
+      core.setSecret(apiKey);
+    }
+
+    // Build config overrides
+    const configOverrides: Partial<EvalConfig> = {
+      defaultAgentModel: model,
+      defaultJudgeModel: judgeModel,
+      discoveryThreshold: thresholdDiscovery,
+      scoreThreshold: thresholdScore,
+      taskTimeoutMs: timeout,
+      githubSummary: true,
+    };
+
+    // Run pipeline
+    const result = await runPipeline({
+      tasksFile: tasks,
+      configPath,
+      configOverrides,
+      cwd,
+      skillsDir,
+      taskFilter: tasksFilter,
+      noJudge,
+      noDeterministic,
+    });
+
+    // Set outputs
+    core.setOutput('passed', String(result.passed));
+    core.setOutput('discovery-rate', String(result.report.summary.discoveryAccuracy));
+    core.setOutput('avg-score', String(result.report.summary.avgWeightedScore));
+    core.setOutput('report-path', result.reportPath || '');
+    core.setOutput('json-path', result.jsonPath || '');
+
+    // Write job summary
+    await core.summary.addRaw(result.markdownSummary).write();
+
+    // Set exit status
+    if (!result.passed) {
+      core.setFailed(
+        `Evaluation below thresholds: ${result.failureReasons.join(', ')}`
+      );
+    }
+  } catch (error) {
+    core.setFailed(error instanceof Error ? error.message : String(error));
+  }
+}
+
+run();

package/dist/action/index.d.ts

@@ -0,0 +1,8 @@
+/**
+ * GitHub Action entry point for skill evaluation.
+ *
+ * Reads inputs from the action.yml, runs the evaluation pipeline,
+ * and sets outputs + job summary.
+ */
+export {};
+//# sourceMappingURL=index.d.ts.map

package/dist/action/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../action/index.ts"],"names":[],"mappings":"AAAA;;;;;GAKG"}

package/dist/action/index.js

@@ -0,0 +1,68 @@
+/**
+ * GitHub Action entry point for skill evaluation.
+ *
+ * Reads inputs from the action.yml, runs the evaluation pipeline,
+ * and sets outputs + job summary.
+ */
+import * as core from '@actions/core';
+import { runPipeline } from '../src/pipeline.js';
+async function run() {
+    try {
+        // Read inputs
+        const tasks = core.getInput('tasks', { required: true });
+        const model = core.getInput('model') || 'sonnet';
+        const judgeModel = core.getInput('judge-model') || 'haiku';
+        const configPath = core.getInput('config') || undefined;
+        const thresholdDiscovery = parseFloat(core.getInput('threshold-discovery') || '0.8');
+        const thresholdScore = parseFloat(core.getInput('threshold-score') || '4.0');
+        const timeout = parseInt(core.getInput('timeout') || '300000', 10);
+        const tasksFilter = core.getInput('tasks-filter') || undefined;
+        const skillsDir = core.getInput('skills-dir') || undefined;
+        const cwd = core.getInput('working-directory') || process.cwd();
+        const noJudge = core.getInput('no-judge') === 'true';
+        const noDeterministic = core.getInput('no-deterministic') === 'true';
+        // Handle API key
+        const apiKey = core.getInput('anthropic-api-key') || process.env.ANTHROPIC_API_KEY;
+        if (apiKey) {
+            process.env.ANTHROPIC_API_KEY = apiKey;
+            core.setSecret(apiKey);
+        }
+        // Build config overrides
+        const configOverrides = {
+            defaultAgentModel: model,
+            defaultJudgeModel: judgeModel,
+            discoveryThreshold: thresholdDiscovery,
+            scoreThreshold: thresholdScore,
+            taskTimeoutMs: timeout,
+            githubSummary: true,
+        };
+        // Run pipeline
+        const result = await runPipeline({
+            tasksFile: tasks,
+            configPath,
+            configOverrides,
+            cwd,
+            skillsDir,
+            taskFilter: tasksFilter,
+            noJudge,
+            noDeterministic,
+        });
+        // Set outputs
+        core.setOutput('passed', String(result.passed));
+        core.setOutput('discovery-rate', String(result.report.summary.discoveryAccuracy));
+        core.setOutput('avg-score', String(result.report.summary.avgWeightedScore));
+        core.setOutput('report-path', result.reportPath || '');
+        core.setOutput('json-path', result.jsonPath || '');
+        // Write job summary
+        await core.summary.addRaw(result.markdownSummary).write();
+        // Set exit status
+        if (!result.passed) {
+            core.setFailed(`Evaluation below thresholds: ${result.failureReasons.join(', ')}`);
+        }
+    }
+    catch (error) {
+        core.setFailed(error instanceof Error ? error.message : String(error));
+    }
+}
+run();
+//# sourceMappingURL=index.js.map

package/dist/action/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../action/index.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,IAAI,MAAM,eAAe,CAAC;AACtC,OAAO,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AAGjD,KAAK,UAAU,GAAG;IAChB,IAAI,CAAC;QACH,cAAc;QACd,MAAM,KAAK,GAAG,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC;QACzD,MAAM,KAAK,GAAG,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC,IAAI,QAAQ,CAAC;QACjD,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC,IAAI,OAAO,CAAC;QAC3D,MAAM,UAAU,GAAG,IAAI,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,SAAS,CAAC;QACxD,MAAM,kBAAkB,GAAG,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,qBAAqB,CAAC,IAAI,KAAK,CAAC,CAAC;QACrF,MAAM,cAAc,GAAG,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,iBAAiB,CAAC,IAAI,KAAK,CAAC,CAAC;QAC7E,MAAM,OAAO,GAAG,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,IAAI,QAAQ,EAAE,EAAE,CAAC,CAAC;QACnE,MAAM,WAAW,GAAG,IAAI,CAAC,QAAQ,CAAC,cAAc,CAAC,IAAI,SAAS,CAAC;QAC/D,MAAM,SAAS,GAAG,IAAI,CAAC,QAAQ,CAAC,YAAY,CAAC,IAAI,SAAS,CAAC;QAC3D,MAAM,GAAG,GAAG,IAAI,CAAC,QAAQ,CAAC,mBAAmB,CAAC,IAAI,OAAO,CAAC,GAAG,EAAE,CAAC;QAChE,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,KAAK,MAAM,CAAC;QACrD,MAAM,eAAe,GAAG,IAAI,CAAC,QAAQ,CAAC,kBAAkB,CAAC,KAAK,MAAM,CAAC;QAErE,iBAAiB;QACjB,MAAM,MAAM,GAAG,IAAI,CAAC,QAAQ,CAAC,mBAAmB,CAAC,IAAI,OAAO,CAAC,GAAG,CAAC,iBAAiB,CAAC;QACnF,IAAI,MAAM,EAAE,CAAC;YACX,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,MAAM,CAAC;YACvC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;QACzB,CAAC;QAED,yBAAyB;QACzB,MAAM,eAAe,GAAwB;YAC3C,iBAAiB,EAAE,KAAK;YACxB,iBAAiB,EAAE,UAAU;YAC7B,kBAAkB,EAAE,kBAAkB;YACtC,cAAc,EAAE,cAAc;YAC9B,aAAa,EAAE,OAAO;YACtB,aAAa,EAAE,IAAI;SACpB,CAAC;QAEF,eAAe;QACf,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC;YAC/B,SAAS,EAAE,KAAK;YAChB,UAAU;YACV,eAAe;YACf,GAAG;YACH,SAAS;YACT,UAAU,EAAE,WAAW;YACvB,OAAO;YACP,eAAe;SAChB,CAAC,CAAC;QAEH,cAAc;QACd,IAAI,CAAC,SAAS,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC;QAChD,IAAI,CAAC,SAAS,CAAC,gBAAgB,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,iBAAiB,CAAC,CAAC,CAAC;QAClF,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC,CAAC;QAC5E,IAAI,CAAC,SAAS,CAAC,aAAa,EAAE,MAAM,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC;QACvD,IAAI,CAAC,SAAS,CAAC,WAAW,EAAE,MAAM,CAAC,QAAQ,IAAI,EAAE,CAAC,CAAC;QAEnD,oBAAoB;QACpB,MAAM,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,CAAC,KAAK,EAAE,CAAC;QAE1D,kBAAkB;QAClB,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;YACnB,IAAI,CAAC,SAAS,CACZ,gCAAgC,MAAM,CAAC,cAAc,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CACnE,CAAC;QACJ,CAAC;IACH,CAAC;IAAC,OAAO,KAAK,EAAE,CAAC;QACf,IAAI,CAAC,SAAS,CAAC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC;IACzE,CAAC;AACH,CAAC;AAED,GAAG,EAAE,CAAC"}

package/dist/src/cli.d.ts

@@ -0,0 +1,9 @@
+#!/usr/bin/env node
+/**
+ * CLI for skill evaluation framework.
+ *
+ * Primary command: `skilljack-evals run` — runs the full evaluation pipeline.
+ * Also supports: score, report, create-eval, validate.
+ */
+import 'dotenv/config';
+//# sourceMappingURL=cli.d.ts.map

package/dist/src/cli.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"cli.d.ts","sourceRoot":"","sources":["../../src/cli.ts"],"names":[],"mappings":";AAEA;;;;;GAKG;AAEH,OAAO,eAAe,CAAC"}