@rely-ai/caliber 1.18.2 → 1.18.3

package/dist/bin.js

@@ -1759,31 +1759,21 @@ async function validateModel(options) {
 init_config();
 
 // src/ai/prompts.ts
-var GENERATION_SYSTEM_PROMPT = `You are an expert auditor for coding agent configurations (Claude Code, Cursor, and Codex).
+var ROLE_AND_CONTEXT = `You are an expert auditor for coding agent configurations (Claude Code, Cursor, and Codex).
 
 Your job depends on context:
 - If no existing configs exist \u2192 generate an initial setup from scratch.
-- If existing configs are provided \u2192 audit them and suggest targeted improvements. Preserve accurate content \u2014 don't rewrite what's already correct.
-
-You understand these config files:
+- If existing configs are provided \u2192 audit them and suggest targeted improvements. Preserve accurate content \u2014 don't rewrite what's already correct.`;
+var CONFIG_FILE_TYPES = `You understand these config files:
 - CLAUDE.md: Project context for Claude Code \u2014 build/test commands, architecture, conventions.
 - AGENTS.md: Primary instructions file for OpenAI Codex \u2014 same purpose as CLAUDE.md but for the Codex agent. Also serves as a cross-agent coordination file.
 - .claude/skills/{name}/SKILL.md: Skill files following the OpenSkills standard (agentskills.io). Each skill is a directory named after the skill, containing a SKILL.md with YAML frontmatter.
 - .agents/skills/{name}/SKILL.md: Same OpenSkills format for Codex skills (Codex scans .agents/skills/ for skills).
+- .cursor/skills/{name}/SKILL.md: Same OpenSkills format for Cursor skills.
 - .cursorrules: Coding rules for Cursor (deprecated legacy format \u2014 do NOT generate this).
-- .cursor/rules/*.mdc: Modern Cursor rules with frontmatter (description, globs, alwaysApply).
-- .cursor/skills/{name}/SKILL.md: Same OpenSkills format as Claude skills.
-
-Audit checklist (when existing configs are provided):
-1. CLAUDE.md / README accuracy \u2014 do documented commands, paths, and architecture match the actual codebase?
-2. Missing skills \u2014 are there detected tools/frameworks that should have dedicated skills?
-3. Duplicate or overlapping skills \u2014 can any be merged or removed?
-4. Undocumented conventions \u2014 are there code patterns (commit style, async patterns, error handling) not captured in docs?
-5. Stale references \u2014 do docs mention removed files, renamed commands, or outdated patterns?
-
-Do NOT generate .claude/settings.json or .claude/settings.local.json \u2014 those are managed by the user directly.
-
-Your output MUST follow this exact format (no markdown fences):
+- .cursor/rules/*.mdc: Modern Cursor rules with frontmatter (description, globs, alwaysApply).`;
+var EXCLUSIONS = `Do NOT generate .claude/settings.json, .claude/settings.local.json, or mcpServers \u2014 those are managed separately.`;
+var OUTPUT_FORMAT = `Your output MUST follow this exact format (no markdown fences):
 
 1. Exactly 6 short status lines (one per line, prefixed with "STATUS: "). Each should be a creative, specific description of what you're analyzing for THIS project \u2014 reference the project's actual languages, frameworks, or tools.
 
@@ -1797,34 +1787,11 @@ EXPLAIN:
 
 Omit empty categories. Keep each reason punchy and specific. End with a blank line.
 
-3. The JSON object starting with {.
-
-AgentSetup schema:
-{
-  "targetAgent": ["claude", "cursor", "codex"] (array of selected agents),
-  "fileDescriptions": {
-    "<file-path>": "reason for this change (max 80 chars)"
-  },
-  "deletions": [
-    { "filePath": "<path>", "reason": "why remove (max 80 chars)" }
-  ],
-  "claude": {
-    "claudeMd": "string (markdown content for CLAUDE.md)",
-    "skills": [{ "name": "string (kebab-case, matches directory name)", "description": "string (what this skill does and when to use it)", "content": "string (markdown body \u2014 NO frontmatter, it will be generated from name+description)" }]
-  },
-  "codex": {
-    "agentsMd": "string (markdown content for AGENTS.md \u2014 the primary Codex instructions file, same quality/structure as CLAUDE.md)",
-    "skills": [{ "name": "string (kebab-case, matches directory name)", "description": "string (what this skill does and when to use it)", "content": "string (markdown body \u2014 NO frontmatter, it will be generated from name+description)" }]
-  },
-  "cursor": {
-    "skills": [{ "name": "string (kebab-case, matches directory name)", "description": "string (what this skill does and when to use it)", "content": "string (markdown body \u2014 NO frontmatter, it will be generated from name+description)" }],
-    "rules": [{ "filename": "string.mdc", "content": "string (with frontmatter)" }]
-  }
-}
-
-Do NOT generate mcpServers \u2014 MCP configuration is managed separately.
+3. The JSON object starting with {.`;
+var FILE_DESCRIPTIONS_RULES = `The "fileDescriptions" object MUST include a one-liner for every file that will be created or modified. Use actual file paths as keys (e.g. "CLAUDE.md", "AGENTS.md", ".claude/skills/my-skill/SKILL.md", ".agents/skills/my-skill/SKILL.md", ".cursor/skills/my-skill/SKILL.md", ".cursor/rules/my-rule.mdc"). Each description should explain why the change is needed, be concise and lowercase.
 
-All skills follow the OpenSkills standard (agentskills.io). Anthropic's official skill guide defines three levels of progressive disclosure:
+The "deletions" array should list files that should be removed (e.g. duplicate skills, stale configs). Include a reason for each. Omit the array or leave empty if nothing should be deleted.`;
+var SKILL_FORMAT_RULES = `All skills follow the OpenSkills standard (agentskills.io). Anthropic's official skill guide defines three levels of progressive disclosure:
 - Level 1 (YAML frontmatter): Always loaded. Must have enough info for the agent to decide when to activate the skill.
 - Level 2 (SKILL.md body): Loaded when the skill is relevant. Contains full instructions.
 - Level 3 (references/): Only loaded on demand for deep detail.
@@ -1840,84 +1807,93 @@ Skill content structure \u2014 follow this template:
 3. "## Examples" \u2014 at least one example showing: User says \u2192 Actions taken \u2192 Result
 4. "## Troubleshooting" (optional) \u2014 common errors and how to fix them
 
-Keep skill content under 200 lines. Focus on actionable instructions, not documentation prose.
-
-The "fileDescriptions" object MUST include a one-liner for every file that will be created or modified. Use actual file paths as keys (e.g. "CLAUDE.md", "AGENTS.md", ".claude/skills/my-skill/SKILL.md", ".agents/skills/my-skill/SKILL.md", ".cursor/skills/my-skill/SKILL.md", ".cursor/rules/my-rule.mdc"). Each description should explain why the change is needed, be concise and lowercase.
-
-The "deletions" array should list files that should be removed (e.g. duplicate skills, stale configs). Include a reason for each. Omit the array or leave empty if nothing should be deleted.
-
-SCORING CRITERIA \u2014 your output is scored deterministically. Optimize for 100/100:
+Keep skill content under 200 lines. Focus on actionable instructions, not documentation prose.`;
+var SCORING_CRITERIA = `SCORING CRITERIA \u2014 your output is scored deterministically against the actual filesystem. Optimize for 100/100:
 
 Existence (25 pts):
-- CLAUDE.md exists (6 pts) \u2014 always generate for claude/both targets
-- AGENTS.md exists (6 pts) \u2014 always generate for codex target (serves as primary instructions file)
-- Skills configured (8 pts) \u2014 generate at least 3 skills for full points. Generate more if the project has multiple distinct tools, frameworks, or workflows that benefit from dedicated skills.
-- MCP servers mentioned (3 pts) \u2014 reference detected MCP integrations
-- For "both" target: .cursorrules/.cursor/rules/ exist (3+3 pts), cross-platform parity (2 pts)
+- CLAUDE.md exists (6 pts) \u2014 always generate for claude targets
+- AGENTS.md exists (6 pts) \u2014 always generate for codex target
+- Skills configured (8 pts) \u2014 generate 3+ skills for full points
+- MCP servers referenced (3 pts) \u2014 mention detected MCP integrations in your config text
+- When cursor is targeted: Cursor rules exist (3+3 pts), cross-platform parity (2 pts)
 
 Quality (25 pts):
-- Build/test/lint commands documented (8 pts) \u2014 include actual commands from the project
-- Concise context files (6 pts) \u2014 keep CLAUDE.md under 150 lines for full points (200=4pts, 300=3pts, 500+=0pts)
-- No vague instructions (4 pts) \u2014 avoid "follow best practices", "write clean code", "ensure quality"
+- Executable content (8 pts) \u2014 include 3+ code blocks with actual project commands (3 blocks = full points, 2 = 6pts, 1 = 3pts)
+- Concise config (6 pts) \u2014 total tokens across ALL config files must be under 2000 for full points (3500=5pts, 5000=4pts, 8000+=low)
+- Concrete instructions (4 pts) \u2014 every line should reference specific files, paths, or code in backticks. Avoid generic prose like "follow best practices" or "write clean code".
 - No directory tree listings (3 pts) \u2014 do NOT include tree-style file listings in code blocks
-- No contradictions (2 pts) \u2014 consistent tool/style recommendations
-
-Coverage (20 pts):
-- Dependency coverage (10 pts) \u2014 CRITICAL: the exact dependency list is provided in your input under "DEPENDENCY COVERAGE". Mention AT LEAST 85% of them by name in CLAUDE.md or skills. You get full points at 85%+, proportional below that. Weave them naturally into architecture, key deps, and conventions sections.
-- Service/MCP coverage (6 pts) \u2014 reference detected services (DB, cloud, etc.)
-- MCP completeness (4 pts) \u2014 full points if no external services detected
-
-Accuracy (15 pts) \u2014 THIS IS CRITICAL, READ CAREFULLY:
-- Documented commands exist (6 pts) \u2014 the scoring system validates EVERY command you write against the project's actual package.json scripts, Makefile targets, or Cargo.toml. If you write "yarn build" but there is no "build" script in package.json, you LOSE points. Rules:
-  * Look at the "scripts" section in the provided package.json. ONLY reference scripts that exist there.
-  * If a project uses Makefiles, only reference targets that exist in the Makefile.
-  * If there are no build/test scripts, do NOT invent them. Document what actually exists.
-  * Use the exact package manager the project uses (npm/yarn/pnpm/bun) \u2014 check the lockfile.
-- Documented paths exist (4 pts) \u2014 ONLY reference file paths from the provided file tree. Never guess paths.
-- Config freshness (5 pts) \u2014 config must match current code state
-
-Freshness & Safety (10 pts):
-- No secrets in configs (4 pts) \u2014 never include API keys, tokens, or credentials
-- Permissions configured (2 pts) \u2014 handled by caliber, not your responsibility
-
-Bonus (5 pts):
-- Hooks configured (2 pts), AGENTS.md (1 pt), OpenSkills format (2 pts) \u2014 handled by caliber
-
-OUTPUT SIZE CONSTRAINTS \u2014 these are critical:
+- No duplicate content (2 pts) \u2014 don't repeat the same content across CLAUDE.md and cursor rules
+- Structured with headings (2 pts) \u2014 use at least 3 ## sections and bullet lists
+
+Grounding (20 pts) \u2014 CRITICAL:
+- Project grounding (12 pts) \u2014 reference the project's actual directories and files by name. The scoring checks which project dirs/files from the file tree appear in your config. Mention key directories and files. (50%+ coverage = full points, 35% = 9pts, 20% = 6pts, 10% = 3pts)
+- Reference density (8 pts) \u2014 use backticks and inline code extensively. Every file path, command, or identifier should be in backticks. Higher density of backtick references per line = higher score. (40%+ = full, 25% = 6pts, 15% = 4pts)
+
+Accuracy (15 pts) \u2014 CRITICAL:
+- References valid (8 pts) \u2014 ONLY reference file paths that exist in the provided file tree. Every path in backticks is validated against the filesystem. If you write a path that doesn't exist, you LOSE points.
+- Config drift (7 pts) \u2014 handled automatically by caliber (git-based), not your responsibility.
+
+Safety: Never include API keys, tokens, or credentials in config files.
+
+Note: Permissions, hooks, freshness tracking, and OpenSkills frontmatter are scored automatically by caliber \u2014 do not optimize for them.`;
+var OUTPUT_SIZE_CONSTRAINTS = `OUTPUT SIZE CONSTRAINTS \u2014 these are critical:
 - CLAUDE.md / AGENTS.md: MUST be under 150 lines for maximum score. Aim for 100-140 lines. Be concise \u2014 commands, architecture overview, and key conventions. Use bullet points and tables, not prose.
-- Skills: generate 3-6 skills per target platform based on project complexity. Each skill should cover a distinct tool, workflow, or domain \u2014 don't pad with generic skills.
 - Each skill content: max 150 lines. Focus on patterns and examples, not exhaustive docs.
 - Cursor rules: max 5 .mdc files.
 - If the project is large, prioritize depth on the 3-4 most critical tools over breadth across everything.`;
-var CORE_GENERATION_PROMPT = `You are an expert auditor for coding agent configurations (Claude Code, Cursor, and Codex).
+var GENERATION_SYSTEM_PROMPT = `${ROLE_AND_CONTEXT}
 
-Your job depends on context:
-- If no existing configs exist \u2192 generate an initial setup from scratch.
-- If existing configs are provided \u2192 audit them and suggest targeted improvements. Preserve accurate content \u2014 don't rewrite what's already correct.
+${CONFIG_FILE_TYPES}
 
-You understand these config files:
-- CLAUDE.md: Project context for Claude Code \u2014 build/test commands, architecture, conventions.
-- AGENTS.md: Primary instructions file for OpenAI Codex \u2014 same purpose as CLAUDE.md but for the Codex agent.
-- .cursorrules: Coding rules for Cursor (deprecated legacy format \u2014 do NOT generate this).
-- .cursor/rules/*.mdc: Modern Cursor rules with frontmatter (description, globs, alwaysApply).
+Audit checklist (when existing configs are provided):
+1. CLAUDE.md / README accuracy \u2014 do documented commands, paths, and architecture match the actual codebase?
+2. Missing skills \u2014 are there detected tools/frameworks that should have dedicated skills?
+3. Duplicate or overlapping skills \u2014 can any be merged or removed?
+4. Undocumented conventions \u2014 are there code patterns (commit style, async patterns, error handling) not captured in docs?
+5. Stale references \u2014 do docs mention removed files, renamed commands, or outdated patterns?
 
-Do NOT generate .claude/settings.json, .claude/settings.local.json, or mcpServers.
+${EXCLUSIONS}
 
-Your output MUST follow this exact format (no markdown fences):
+${OUTPUT_FORMAT}
 
-1. Exactly 6 short status lines (one per line, prefixed with "STATUS: "). Each should be a creative, specific description of what you're analyzing for THIS project \u2014 reference the project's actual languages, frameworks, or tools.
+AgentSetup schema:
+{
+  "targetAgent": ["claude", "cursor", "codex"] (array of selected agents),
+  "fileDescriptions": {
+    "<file-path>": "reason for this change (max 80 chars)"
+  },
+  "deletions": [
+    { "filePath": "<path>", "reason": "why remove (max 80 chars)" }
+  ],
+  "claude": {
+    "claudeMd": "string (markdown content for CLAUDE.md)",
+    "skills": [{ "name": "string (kebab-case, matches directory name)", "description": "string (what this skill does and when to use it)", "content": "string (markdown body \u2014 NO frontmatter, it will be generated from name+description)" }]
+  },
+  "codex": {
+    "agentsMd": "string (markdown content for AGENTS.md \u2014 the primary Codex instructions file, same quality/structure as CLAUDE.md)",
+    "skills": [{ "name": "string (kebab-case, matches directory name)", "description": "string (what this skill does and when to use it)", "content": "string (markdown body \u2014 NO frontmatter, it will be generated from name+description)" }]
+  },
+  "cursor": {
+    "skills": [{ "name": "string (kebab-case, matches directory name)", "description": "string (what this skill does and when to use it)", "content": "string (markdown body \u2014 NO frontmatter, it will be generated from name+description)" }],
+    "rules": [{ "filename": "string.mdc", "content": "string (with frontmatter)" }]
+  }
+}
 
-2. A brief explanation section starting with "EXPLAIN:" on its own line:
+${SKILL_FORMAT_RULES}
 
-EXPLAIN:
-[Changes]
-- **file-or-skill-name**: short reason (max 10 words)
-[Deletions]
-- **file-path**: short reason (max 10 words)
+${FILE_DESCRIPTIONS_RULES}
 
-Omit empty categories. Keep each reason punchy and specific. End with a blank line.
+${SCORING_CRITERIA}
+
+${OUTPUT_SIZE_CONSTRAINTS}
+- Skills: generate 3-6 skills per target platform based on project complexity. Each skill should cover a distinct tool, workflow, or domain \u2014 don't pad with generic skills.`;
+var CORE_GENERATION_PROMPT = `${ROLE_AND_CONTEXT}
+
+${CONFIG_FILE_TYPES}
 
-3. The JSON object starting with {.
+${EXCLUSIONS}
+
+${OUTPUT_FORMAT}
 
 CoreSetup schema:
 {
@@ -1955,40 +1931,11 @@ Skill topic description MUST follow this formula: [What it does] + [When to use
 Include specific trigger phrases users would actually say. Also include negative triggers to prevent over-triggering.
 Example: "Creates a new API endpoint following the project's route pattern. Handles request validation, error responses, and DB queries. Use when user says 'add endpoint', 'new route', 'create API', or adds files to src/routes/. Do NOT use for modifying existing routes."
 
-The "fileDescriptions" object MUST include a one-liner for every file that will be created or modified.
-The "deletions" array should list files that should be removed (e.g. stale configs). Omit if empty.
+${FILE_DESCRIPTIONS_RULES}
 
-SCORING CRITERIA \u2014 your output is scored deterministically against the actual filesystem. Optimize for 100/100:
+${SCORING_CRITERIA}
 
-Existence (25 pts):
-- CLAUDE.md exists (6 pts) \u2014 always generate for claude targets
-- AGENTS.md exists (6 pts) \u2014 always generate for codex target
-- Skills configured (8 pts) \u2014 generate 3+ skill topics for full points
-- For "both" target: .cursor/rules/ exist (3+3 pts), cross-platform parity (2 pts)
-
-Quality (25 pts):
-- Executable content (8 pts) \u2014 include 3+ code blocks with project commands (3 blocks = full points)
-- Concise config (6 pts) \u2014 total tokens across ALL config files must be under 2000 for full points (5000=4pts, 8000+=low)
-- Concrete instructions (4 pts) \u2014 every line should reference specific files, paths, or code in backticks. Avoid generic prose.
-- No directory tree listings (3 pts) \u2014 do NOT include tree-style file listings
-- Structured with headings (2 pts) \u2014 use at least 3 ## sections and bullet lists
-
-Grounding (20 pts) \u2014 CRITICAL:
-- Project grounding (12 pts) \u2014 reference the project's actual directories and files by name. The scoring checks which project dirs/files appear in your config. Mention key directories from the file tree.
-- Reference density (8 pts) \u2014 use backticks and inline code extensively. Every file path, command, or identifier should be in backticks. Higher density of specific references = higher score.
-
-Accuracy (15 pts) \u2014 CRITICAL:
-- References valid (8 pts) \u2014 ONLY reference file paths that exist in the provided file tree. Every path in backticks is validated against the filesystem.
-- Config drift (7 pts) \u2014 config must match current code state
-
-Freshness & Safety (10 pts):
-- No secrets (4 pts), Permissions (2 pts \u2014 handled by caliber), Freshness (4 pts \u2014 handled by caliber)
-
-Bonus (5 pts): Hooks (2 pts), AGENTS.md (1 pt), OpenSkills format (2 pts) \u2014 handled by caliber
-
-OUTPUT SIZE CONSTRAINTS:
-- CLAUDE.md / AGENTS.md: MUST be under 150 lines. Aim for 100-140 lines.
-- Cursor rules: max 5 .mdc files.
+${OUTPUT_SIZE_CONSTRAINTS}
 - Skill topics: 3-6 per platform based on project complexity (name + description only, no content).`;
 var SKILL_GENERATION_PROMPT = `You generate a single skill file for a coding agent (Claude Code, Cursor, or Codex).
 
@@ -2053,12 +2000,20 @@ Rules:
 - Preserve all fields that the user did not ask to change.
 - Do NOT generate mcpServers \u2014 MCP configuration is managed separately.
 - Skills use OpenSkills format: name is kebab-case directory name, content is markdown body without frontmatter.
-- Update the "fileDescriptions" to reflect any changes you make.`;
+- Update the "fileDescriptions" to reflect any changes you make.
+
+Quality constraints \u2014 your changes are scored, so do not break these:
+- CLAUDE.md / AGENTS.md: MUST stay under 150 lines. If adding content, remove less important lines to stay within budget.
+- Avoid vague instructions ("follow best practices", "write clean code", "ensure quality").
+- Do NOT add directory tree listings in code blocks.
+- Use backticks for every file path, command, and identifier.
+- Keep skill content under 150 lines, focused on actionable instructions.
+- Only reference file paths that actually exist in the project.`;
 var REFRESH_SYSTEM_PROMPT = `You are an expert at maintaining coding project documentation. Your job is to update existing documentation files based on code changes (git diffs).
 
 You will receive:
 1. Git diffs showing what code changed
-2. Current contents of documentation files (CLAUDE.md, AGENTS.md, README.md, skills, cursor skills, cursor rules)
+2. Current contents of documentation files (CLAUDE.md, README.md, skills, cursor rules)
 3. Project context (languages, frameworks)
 
 Rules:
@@ -2068,23 +2023,22 @@ Rules:
 - Don't add speculative or aspirational content
 - Keep managed blocks (<!-- caliber:managed --> ... <!-- /caliber:managed -->) intact
 - If a doc doesn't need updating, return null for it
-- For CLAUDE.md: update commands, architecture notes, conventions, key files if the diffs affect them
-- For AGENTS.md: same as CLAUDE.md \u2014 this is the primary instructions file for Codex users
+- For CLAUDE.md: update commands, architecture notes, conventions, key files if the diffs affect them. Keep under 150 lines.
 - For README.md: update setup instructions, API docs, or feature descriptions if affected
-- For cursor skills: update skill content if the diffs affect their domains
+- Only reference file paths that exist in the project
+- Use backticks for all file paths, commands, and identifiers
 
 Return a JSON object with this exact shape:
 {
   "updatedDocs": {
     "claudeMd": "<updated content or null>",
-    "agentsMd": "<updated content or null>",
     "readmeMd": "<updated content or null>",
+    "cursorrules": "<updated content or null>",
     "cursorRules": [{"filename": "name.mdc", "content": "..."}] or null,
-    "cursorSkills": [{"slug": "string", "name": "string", "content": "..."}] or null,
     "claudeSkills": [{"filename": "name.md", "content": "..."}] or null
   },
   "changesSummary": "<1-2 sentence summary of what was updated and why>",
-  "docsUpdated": ["CLAUDE.md", "AGENTS.md", "README.md"]
+  "docsUpdated": ["CLAUDE.md", "README.md"]
 }
 
 Respond with ONLY the JSON object, no markdown fences or extra text.`;
@@ -2113,7 +2067,7 @@ Rules for the learned section:
 - Be additive: keep all existing learned items, add new ones, remove duplicates
 - Never repeat instructions already present in the main CLAUDE.md
 - Each bullet must be specific and actionable \u2014 no vague advice
-- Maximum ~50 bullet items total
+- Maximum ~30 bullet items total
 - Group related items under subheadings if there are many
 - If there's nothing meaningful to learn, return null
 
@@ -2148,7 +2102,7 @@ Be thorough \u2014 look for signals in:
 - Configuration files (e.g. next.config.js implies Next.js, .tf files imply Terraform + cloud providers)
 - Infrastructure-as-code files (Terraform, CloudFormation, Pulumi, Dockerfiles, k8s manifests)
 - CI/CD configs (.github/workflows, .gitlab-ci.yml, Jenkinsfile)
-- Environment variable patterns and service references in code
+- Dockerfile base images (e.g. FROM python:3.11 implies Python, FROM node:20 implies Node.js)
 
 Only include items you're confident about. Return ONLY the JSON object.`;
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@rely-ai/caliber",
-  "version": "1.18.2",
+  "version": "1.18.3",
   "description": "Analyze your codebase and generate optimized AI agent configs (CLAUDE.md, .cursorrules, skills) — no API key needed",
   "type": "module",
   "bin": {