@a5c-ai/tula 5.0.1-staging.3edc9d7b8d6f

package/dist/prompts/commandTemplates.js

@@ -0,0 +1,238 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.renderCommandTemplate = renderCommandTemplate;
+const babysitter_sdk_1 = require("@a5c-ai/babysitter-sdk");
+const COMMAND_TEMPLATE_CONTEXT = {
+    harness: "harness-command-template",
+    harnessLabel: "Harness Command Template",
+    interactive: undefined,
+    capabilities: [],
+    platform: process.platform,
+    pluginRootVar: "",
+    loopControlTerm: "",
+    sessionBindingFlags: "",
+    hookDriven: false,
+    interactiveToolName: "",
+    sessionEnvVars: "",
+    resumeFlags: "",
+    sdkVersionExpr: "",
+    hasIntentFidelityChecks: false,
+    hasNonNegotiables: false,
+    cliSetupSnippet: "",
+    iterateFlags: "",
+};
+const ANYCLI_TEMPLATE = `[ANYCLI MODE] Dynamic service integration agent for any external service
+
+Target service: {{serviceName}}
+Scope: {{scope}}
+{{#mcpMode}}MCP mode: enabled (transport: {{transport}}){{/mcpMode}}
+{{^mcpMode}}Mode: ad-hoc execution{{/mcpMode}}
+{{#authFile}}Auth file: {{authFile}}{{/authFile}}
+{{#userPrompt}}User request: {{userPrompt}}{{/userPrompt}}
+
+You are a service integration agent. You handle EVERYTHING dynamically: researching the service, figuring out authentication, writing code, and generating MCP servers. There are no pre-built templates or hardcoded patterns -- you discover and create everything at runtime based on what you learn about the service.
+
+---
+
+## Phase 1: Service Discovery
+
+**Goal:** Learn everything about **{{serviceName}}** and build a service definition.
+
+### 1a. Check Cache
+
+Check \`~/.a5c/anycli/cache/{{serviceName}}/cache.json\` for an existing cache entry.
+
+If found, inspect \`definition\`, \`modules\`, and \`metadata.createdAt\`. You decide whether the cache is fresh enough to use. If the definition covers the requested scope (\`{{scope}}\`), skip to Phase 3 or 4.
+
+### 1b. Research the Service
+
+If no usable cache exists, research **{{serviceName}}** thoroughly:
+
+1. **Web search** for official API documentation, OpenAPI/Swagger specs, developer guides.
+2. **Identify**: base URL(s), API versioning scheme, authentication methods (specific to THIS service -- not generic patterns), rate limits, pagination style, error response format.
+3. **Check** for existing CLI tools, SDKs (npm/pip/etc.), or MCP servers that already integrate with this service.
+4. **Filter by scope**: \`{{scope}}\` -- if \`*\`, cover all major endpoint groups. If comma-separated, focus on those specific areas.
+
+### 1c. Build the Definition
+
+Construct a service definition object with whatever structure makes sense for **{{serviceName}}**. There is no rigid schema -- include the fields that are relevant:
+
+- \`name\`, \`apiBaseUrl\`, \`displayName\`, \`description\`
+- Authentication details specific to this service
+- Endpoint groups filtered by scope
+- Any service-specific metadata (rate limits, pagination, versioning, etc.)
+
+---
+
+## Phase 2: Auth Resolution
+
+**Goal:** Figure out what credentials **{{serviceName}}** needs and find them.
+
+This is entirely service-specific. Do NOT use generic patterns like \`SERVICE_API_KEY\`. Instead:
+
+1. **Read the service docs** to understand what auth method(s) it supports (API keys, OAuth2, bearer tokens, basic auth, etc.) and what the credentials look like.
+{{#authFile}}
+2. **Check the auth file** at \`{{authFile}}\` for credentials. Parse it as \`.env\` or JSON format.
+{{/authFile}}
+{{^authFile}}
+2. **No auth file provided**, skip to next step.
+{{/authFile}}
+3. **Check environment variables** using the naming conventions that are standard for THIS specific service. For example: \`GITHUB_TOKEN\` for GitHub, \`STRIPE_API_KEY\` for Stripe, \`OPENAI_API_KEY\` for OpenAI -- whatever the service's own docs and ecosystem use.
+4. **Check workspace \`.env\`** file for the same service-specific variable names.
+5. **If credentials are missing**, report exactly what is needed, where to get it (with links to the service's credential/API key page if known), and what env var name to use.
+
+**Security:** Never log or hardcode credentials. Generated code must read credentials from environment variables at runtime.
+
+---
+
+## Phase 3: Code Generation
+
+**Goal:** Write utility modules (\`.mjs\` files) based on what you learned about **{{serviceName}}**.
+
+You decide the module structure, function signatures, and types based on the service's actual API. General guidelines:
+
+- **HTTP client**: Use \`fetch()\` only -- zero external dependencies. Include auth headers, retries on 429/5xx, and the service's pagination pattern.
+- **Scope modules**: One file per scope group with async functions for each endpoint. Full JSDoc.
+- **Barrel export**: An \`index.mjs\` re-exporting everything.
+- **No hardcoded credentials**: Read from env vars at runtime.
+
+Write all modules to \`~/.a5c/anycli/cache/{{serviceName}}/modules/\`.
+
+---
+
+{{#mcpMode}}
+## Phase 4: MCP Server Generation
+
+**Goal:** Write and start a complete MCP server for **{{serviceName}}**.
+
+Write a self-contained \`mcp-server.mjs\` file from scratch using \`@modelcontextprotocol/sdk\` patterns:
+
+- Import \`McpServer\` from \`@modelcontextprotocol/sdk/server/mcp.js\`
+- Import the appropriate transport for \`{{transport}}\`:
+  - \`stdio\`: \`StdioServerTransport\` from \`.../server/stdio.js\`
+  - \`http-sse\`: \`SSEServerTransport\` from \`.../server/sse.js\`
+- Register MCP tools based on the scope modules you generated
+- Each tool: descriptive name (\`{{serviceName}}_<action>\`), JSON Schema input, handler that calls the scope module functions
+- Graceful shutdown on SIGINT/SIGTERM
+
+Save to \`~/.a5c/anycli/cache/{{serviceName}}/mcp-server.mjs\` and start it.
+
+Output MCP client config for the user:
+\`\`\`json
+{ "name": "{{serviceName}}-anycli", "transport": "{{transport}}",
+  "command": "node", "args": ["~/.a5c/anycli/cache/{{serviceName}}/mcp-server.mjs"] }
+\`\`\`
+{{/mcpMode}}
+
+{{^mcpMode}}
+## Phase 4: Ad-hoc Execution
+
+**Goal:** Fulfill the user's request using the generated modules.
+
+{{#userPrompt}}
+### Execute request:
+> {{userPrompt}}
+
+1. Determine which API calls are needed.
+2. Import the generated modules from \`~/.a5c/anycli/cache/{{serviceName}}/modules/\`.
+3. Execute operations in logical order, chaining multi-step calls as needed.
+4. Present results clearly: tables for lists, formatted JSON for objects, summaries for aggregations.
+{{/userPrompt}}
+
+{{^userPrompt}}
+### Interactive discovery:
+No specific request provided. List the generated modules and their exported functions with descriptions and parameters. Suggest common use cases for **{{serviceName}}** and ask the user what they want to do.
+{{/userPrompt}}
+{{/mcpMode}}
+
+---
+
+## Phase 5: Caching
+
+**Goal:** Save everything to \`~/.a5c/anycli/cache/{{serviceName}}/cache.json\` for reuse.
+
+Write a single cache entry with this structure:
+\`\`\`json
+{
+  "service": "{{serviceName}}",
+  "definition": { "...your service definition from Phase 1..." },
+  "modules": { "filename.mjs": "...file content..." },
+  "metadata": {
+    "createdAt": "<ISO timestamp>",
+    "sdkVersion": "<version>",
+    "definitionHash": "<hash of the definition>"
+  }
+}
+\`\`\`
+
+The harness package provides \`readServiceCache\`, \`writeServiceCache\`, \`invalidateServiceCache\`, and \`listCachedServices\` functions in \`@a5c-ai/agent-platform\` (anycli module) for this purpose.
+
+---
+
+## Phase 6: Verification
+
+1. **Syntax check**: \`node --check\` on each generated \`.mjs\` file.
+2. **Import check**: Dynamic import of \`modules/index.mjs\` to verify exports.
+3. **Smoke test**: If auth is available, make one lightweight API call (health endpoint or list with limit=1).
+{{#mcpMode}}
+4. **MCP validation**: Verify the server module loads and tool count matches expectations.
+{{/mcpMode}}
+
+---
+
+## Error Handling
+
+- **Unknown service**: Report, suggest alternative names or provide a base URL. Stop.
+- **Missing credentials**: Report what was tried, provide setup instructions with the correct env var names for this service. Stop.
+- **Rate limiting**: Wait if reset < 60s, otherwise report partial results.
+- **Network errors**: Report endpoint URL and error. Suggest checking connectivity.
+- **Security invariant**: Never include credentials in output, errors, logs, or generated comments.
+
+---
+
+## Output Format
+
+\`\`\`
+==============================================
+  ANYCLI SERVICE INTEGRATION REPORT
+  Service: {{serviceName}}    Scope: {{scope}}
+==============================================
+
+OVERALL STATUS: <READY | PARTIAL | FAILED>
+
+| #  | Phase              | Status |
+|----|--------------------|--------|
+| 1  | Service Discovery  | <s>    |
+| 2  | Auth Resolution    | <s>    |
+| 3  | Code Generation    | <s>    |
+| 4  | Execution          | <s>    |
+| 5  | Caching            | <s>    |
+| 6  | Verification       | <s>    |
+
+Cache: ~/.a5c/anycli/cache/{{serviceName}}/
+Modules: <list>    Endpoints: <count>    Auth: <method> (<source>)
+{{#mcpMode}}
+MCP server: ~/.a5c/anycli/cache/{{serviceName}}/mcp-server.mjs
+Transport: {{transport}}
+{{/mcpMode}}
+\`\`\`
+
+**Status:** READY = all phases passed. PARTIAL = warnings but no failures. FAILED = any phase failed.
+{{#mcpMode}}
+Include MCP connection instructions for the user's harness.
+{{/mcpMode}}
+{{^mcpMode}}
+{{#userPrompt}}
+Include execution results prominently before the phase summary.
+{{/userPrompt}}
+{{/mcpMode}}
+`;
+function renderAnycliTemplate(extras) {
+    return (0, babysitter_sdk_1.renderTemplateString)(ANYCLI_TEMPLATE, COMMAND_TEMPLATE_CONTEXT, extras);
+}
+function renderCommandTemplate(templateName, extras) {
+    if (templateName === "anycli") {
+        return renderAnycliTemplate(extras);
+    }
+    return (0, babysitter_sdk_1.renderCommandTemplate)(templateName, extras);
+}

package/package.json

@@ -0,0 +1,76 @@
+{
+  "name": "@a5c-ai/tula",
+  "version": "5.0.1-staging.3edc9d7b8d6f",
+  "description": "Tula — unified agent product integrating core, runtime, platform, and TUI",
+  "license": "MIT",
+  "type": "commonjs",
+  "main": "dist/index.js",
+  "types": "dist/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./dist/index.d.ts",
+      "require": "./dist/index.js",
+      "default": "./dist/index.js"
+    },
+    "./cli": {
+      "types": "./dist/cli/main.d.ts",
+      "require": "./dist/cli/main.js",
+      "default": "./dist/cli/main.js"
+    },
+    "./package.json": "./package.json"
+  },
+  "bin": {
+    "tula": "./dist/cli/main.js"
+  },
+  "files": [
+    "dist/",
+    "README.md"
+  ],
+  "publishConfig": {
+    "access": "public"
+  },
+  "scripts": {
+    "build": "npm run build:runtime --prefix ../.. && tsc --build tsconfig.json",
+    "clean": "rimraf dist tsconfig.tsbuildinfo",
+    "test": "vitest run --config vitest.config.ts"
+  },
+  "dependencies": {
+    "@a5c-ai/tula-core": "5.0.1-staging.3edc9d7b8d6f",
+    "@a5c-ai/agent-runtime": "5.0.1-staging.3edc9d7b8d6f",
+    "@a5c-ai/agent-platform": "5.0.1-staging.3edc9d7b8d6f",
+    "@a5c-ai/agent-mux": "5.0.1-staging.3edc9d7b8d6f",
+    "@a5c-ai/babysitter-sdk": "5.0.1-staging.3edc9d7b8d6f",
+    "@modelcontextprotocol/sdk": "^1.12.1",
+    "@sinclair/typebox": "^0.34.48"
+  },
+  "devDependencies": {
+    "@types/node": "^20.11.30",
+    "rimraf": "^6.1.3",
+    "typescript": "^5.4.2",
+    "vitest": "^4.0.18"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/a5c-ai/babysitter.git",
+    "directory": "packages/tula"
+  },
+  "homepage": "https://github.com/a5c-ai/babysitter/tree/main/packages/tula#readme",
+  "bugs": {
+    "url": "https://github.com/a5c-ai/babysitter/issues"
+  },
+  "keywords": [
+    "babysitter",
+    "tula",
+    "agent",
+    "orchestration",
+    "unified"
+  ],
+  "atlas": {
+    "layers": [
+      "L4",
+      "L5",
+      "L6"
+    ],
+    "muxes": []
+  }
+}