@candor.sh/cli 0.3.0 → 0.3.1

package/dist/index.cjs

@@ -4513,7 +4513,7 @@ function getConfigPath() {
 }
 
 // src/skill-content.ts
-var SKILL_CONTENT = '# Candor \u2014 AI-Moderated User Studies & Human Evaluation\n\nCandor lets you create and manage AI-moderated user studies and human evaluation jobs directly from the terminal. Use `candor study` for product research (AI voice interviews with real users) and `candor eval` for quick human labeling, ranking, and evaluation tasks via Mechanical Turk.\n\n## Available Commands\n\nRun these via the Bash tool. Always use `--json` for machine-readable output.\n\n### Studies (product research)\n\n- `candor study create --url <url> --goal "<goal>" [--participants N] --json` \u2014 Create a study\n- `candor study list --json` \u2014 List all studies (or `candor study --json`)\n- `candor study status <study-id> --json` \u2014 Check participant progress\n- `candor study findings <study-id> --json` \u2014 Get prioritized feedback (P0\u2013P3)\n- `candor study approve <study-id> --json` \u2014 Approve recruitment (human-in-the-loop gate)\n\n### Evals (quick human evaluation / labeling)\n\n- `candor eval create --goal "<goal>" --items "<item1,item2,...>" [--type pairwise_comparison] [--workers N] [--reward <cents>] --json` \u2014 Create an eval\n- `candor eval list --json` \u2014 List all evals\n- `candor eval approve <eval-id> --json` \u2014 Launch eval on MTurk\n- `candor eval status <eval-id> --json` \u2014 Check eval progress\n- `candor eval results <eval-id> --json` \u2014 View results and rankings\n\n## Workflow\n\n### Creating a Study\n\n1. **Infer context** \u2014 If the user doesn\'t provide a URL, check:\n   - `package.json` `homepage` field\n   - README for a deployed URL\n   - Ask the user if nothing is found\n2. **Run `candor study create`** with the product URL and goal\n3. **Show the study summary** including:\n   - Drafted study script (sections, questions, tasks)\n   - Participant count and demographic filters\n   - Estimated cost per session\n4. **Wait for explicit approval** \u2014 Do NOT run `candor study approve` without the user saying "yes", "approve", "go ahead", or similar\n5. **On approval**, run `candor study approve <study-id>` to begin recruiting\n\n### Creating an Eval\n\nEvals are for quick human evaluation tasks like labeling, ranking, rating, or comparing items. They run on Mechanical Turk and return results in minutes.\n\n**When to use eval vs study:**\n- Use `candor eval` when the user wants to label, rank, rate, compare, or categorize items (audio files, images, text, etc.)\n- Use `candor study` when the user wants qualitative product feedback from real users\n\n**Choosing the right task type:**\n- `pairwise_comparison` \u2014 Best for ranking. Workers compare pairs and pick a winner. Produces a stack ranking with win rates. Use when the user says "rank", "compare", "which is better", "stack rank".\n- `categorical_label` \u2014 Workers assign a label from a set of categories. Use when the user says "label", "categorize", "classify", "tag".\n- `rating_scale` \u2014 Workers rate items on a numeric scale. Use when the user says "rate", "score", "quality score".\n- `free_text` \u2014 Workers provide open-ended text feedback.\n\n**Workflow:**\n1. **Identify the items** \u2014 Look for files the user mentions (audio, images, etc.). Extract labels from filenames.\n2. **Infer the task type** \u2014 Based on the user\'s goal, pick the most appropriate type. If vague (e.g. "evaluate these"), default to `pairwise_comparison` for ranking or `rating_scale` for quality assessment.\n3. **Design the eval** \u2014 The API auto-generates the experiment design (randomized pairs, worker counts, etc.) from the goal and items. If the user is prescriptive about methodology, pass their preferences.\n4. **Create the eval** \u2014 Run `candor eval create` with goal and items.\n5. **Show the plan** \u2014 Display task type, item count, pair count, estimated cost and time.\n6. **Wait for approval** \u2014 Do NOT launch without explicit confirmation.\n7. **Launch** \u2014 Run `candor eval approve <id>` to create HITs on MTurk.\n8. **Poll and display results** \u2014 Use `candor eval status <id>` and `candor eval results <id>` to show progress and rankings.\n\n**Example natural language \u2192 eval mapping:**\n\n| User says | Task type | Items |\n|-----------|-----------|-------|\n| "rank these audio samples by TTS quality" | pairwise_comparison | Audio file names |\n| "label these images as cat or dog" | categorical_label | Image file names |\n| "rate these UI mockups on a 1-5 scale" | rating_scale | Mockup file names |\n| "which of these headlines is more engaging" | pairwise_comparison | Headline text |\n| "evaluate these samples" (vague) | pairwise_comparison | File names |\n\n### Checking Results\n\n- Use `candor study status <id> --json` to show completion progress\n- Use `candor study findings <id> --json` to show prioritized findings\n- Use `candor eval status <id> --json` to show eval progress\n- Use `candor eval results <id> --json` to show rankings and agreement metrics\n- Offer to create GitHub issues for P0/P1 items if the user wants\n\n## Important Rules\n\n- NEVER approve a study or launch an eval without explicit human confirmation\n- ALWAYS show cost estimate before launching\n- Studies are billed per participant session (~$14.50/session)\n- Evals are billed per task assignment (~$0.05-0.10/task + 20-40% MTurk fee)\n- If Candor is not initialized, tell the user to run `candor init` first\n';
+var SKILL_CONTENT = '# Candor \u2014 AI-Moderated User Studies & Human Evaluation\n\nCandor lets you create and manage AI-moderated user studies and human evaluation jobs directly from the terminal. Use `candor study` for product research (AI voice interviews with real users) and `candor eval` for quick human labeling, ranking, and evaluation tasks.\n\n## Available Commands\n\nRun these via the Bash tool. Always use `--json` for machine-readable output. Run any command with `--help` for full usage details (e.g. `candor eval create --help`).\n\n### Studies (product research)\n\n- `candor study create --url <url> --goal "<goal>" [--participants N] --json` \u2014 Create a study\n- `candor study list --json` \u2014 List all studies (or `candor study --json`)\n- `candor study status <study-id> --json` \u2014 Check participant progress\n- `candor study findings <study-id> --json` \u2014 Get prioritized feedback (P0\u2013P3)\n- `candor study approve <study-id> --json` \u2014 Approve recruitment (human-in-the-loop gate)\n\n### Evals (quick human evaluation / labeling)\n\n- `candor eval create --goal "<goal>" --items "<item1,item2,...>" [--type pairwise_comparison] [--workers N] [--reward <cents>] --json` \u2014 Create an eval\n- `candor eval list --json` \u2014 List all evals\n- `candor eval approve <eval-id> --json` \u2014 Launch eval\n- `candor eval status <eval-id> [--live] --json` \u2014 Check eval progress (use `--live` for real-time monitoring)\n- `candor eval results <eval-id> --json` \u2014 View results and rankings\n- `candor eval cancel <eval-id> --json` \u2014 Cancel eval and stop recruiting (keeps collected responses)\n\n## Workflow\n\n### Creating a Study\n\n1. **Infer context** \u2014 If the user doesn\'t provide a URL, check:\n   - `package.json` `homepage` field\n   - README for a deployed URL\n   - Ask the user if nothing is found\n2. **Run `candor study create`** with the product URL and goal\n3. **Show the study summary** including:\n   - Drafted study script (sections, questions, tasks)\n   - Participant count and demographic filters\n   - Estimated cost per session\n4. **Wait for explicit approval** \u2014 Do NOT run `candor study approve` without the user saying "yes", "approve", "go ahead", or similar\n5. **On approval**, run `candor study approve <study-id>` to begin recruiting\n\n### Creating an Eval\n\nEvals are for quick human evaluation tasks like labeling, ranking, rating, or comparing items. They return results in minutes.\n\n**When to use eval vs study:**\n- Use `candor eval` when the user wants to label, rank, rate, compare, or categorize items (audio files, images, text, etc.)\n- Use `candor study` when the user wants qualitative product feedback from real users\n\n**Choosing the right task type:**\n- `pairwise_comparison` \u2014 Best for ranking. Workers compare pairs and pick a winner. Produces a stack ranking with win rates. Use when the user says "rank", "compare", "which is better", "stack rank".\n- `categorical_label` \u2014 Workers assign a label from a set of categories. Use when the user says "label", "categorize", "classify", "tag".\n- `rating_scale` \u2014 Workers rate items on a numeric scale. Use when the user says "rate", "score", "quality score".\n- `free_text` \u2014 Workers provide open-ended text feedback.\n\n**Workflow:**\n1. **Identify the items** \u2014 Look for files the user mentions (audio, images, etc.). Extract labels from filenames.\n2. **Infer the task type** \u2014 Based on the user\'s goal, pick the most appropriate type. If vague (e.g. "evaluate these"), default to `pairwise_comparison` for ranking or `rating_scale` for quality assessment.\n3. **Design the eval** \u2014 The API auto-generates the experiment design (randomized pairs, worker counts, etc.) from the goal and items. If the user is prescriptive about methodology, pass their preferences.\n4. **Create the eval** \u2014 Run `candor eval create` with goal and items.\n5. **Show the plan** \u2014 Display task type, item count, pair count, estimated cost and time.\n6. **Wait for approval** \u2014 Do NOT launch without explicit confirmation.\n7. **Launch** \u2014 Run `candor eval approve <id>` to begin recruiting workers.\n8. **Monitor** \u2014 Use `candor eval status <id> --live` for real-time progress, or poll with `candor eval status <id>`.\n9. **Results** \u2014 Use `candor eval results <id>` to show rankings and agreement metrics.\n10. **Cancel** (if needed) \u2014 Use `candor eval cancel <id>` to stop recruiting while keeping collected responses.\n\n**Example natural language \u2192 eval mapping:**\n\n| User says | Task type | Items |\n|-----------|-----------|-------|\n| "rank these audio samples by TTS quality" | pairwise_comparison | Audio file names |\n| "label these images as cat or dog" | categorical_label | Image file names |\n| "rate these UI mockups on a 1-5 scale" | rating_scale | Mockup file names |\n| "which of these headlines is more engaging" | pairwise_comparison | Headline text |\n| "evaluate these samples" (vague) | pairwise_comparison | File names |\n\n### Checking Results\n\n- Use `candor study status <id> --json` to show completion progress\n- Use `candor study findings <id> --json` to show prioritized findings\n- Use `candor eval status <id> --json` to show eval progress\n- Use `candor eval results <id> --json` to show rankings and agreement metrics\n- Offer to create GitHub issues for P0/P1 items if the user wants\n\n## Important Rules\n\n- NEVER approve a study or launch an eval without explicit human confirmation\n- ALWAYS show cost estimate before launching\n- Studies are billed per participant session (~$14.50/session)\n- Evals are billed per task assignment (~$0.05-0.15/task)\n- If Candor is not initialized, tell the user to run `candor init` first\n';
 
 // src/commands/init.ts
 var CLAUDE_SKILLS_DIR = (0, import_path2.join)((0, import_os2.homedir)(), ".claude", "skills", "candor");
@@ -4970,8 +4970,8 @@ async function updateCommand() {
   }
 }
 function getCurrentVersion() {
-  if ("0.3.0") {
-    return "0.3.0";
+  if ("0.3.1") {
+    return "0.3.1";
   }
   try {
     const pkgPath = (0, import_path4.join)((0, import_path4.dirname)(new URL(import_meta2.url).pathname), "..", "package.json");
@@ -5383,7 +5383,7 @@ async function evalCancelCommand(id, options) {
 
 // src/index.ts
 var program2 = new Command();
-program2.name("candor").description("AI-moderated user studies, wired into your dev workflow").version("0.3.0").enablePositionalOptions();
+program2.name("candor").description("AI-moderated user studies, wired into your dev workflow").version("0.3.1").enablePositionalOptions();
 program2.command("init").description("Set up Candor: authenticate and configure Claude Code integration").option("--skill-only", "Only reinstall the skill file").action(initCommand);
 program2.command("login").description("Re-authenticate with Candor").action(loginCommand);
 program2.command("logout").description("Log out and clear stored credentials").action(logoutCommand);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@candor.sh/cli",
-  "version": "0.3.0",
+  "version": "0.3.1",
   "private": false,
   "type": "module",
   "bin": {