palmier 0.6.8 → 0.6.9

package/test/agent-instructions.test.ts

@@ -3,18 +3,57 @@ import assert from "node:assert/strict";
 import * as fs from "fs";
 import * as path from "path";
 import { fileURLToPath } from "url";
-import { generateEndpointDocs, agentTools } from "../src/mcp-tools.js";
+import { generateEndpointDocs, type ToolDefinition } from "../src/mcp-tools.js";
 
 const __dirname = path.dirname(fileURLToPath(import.meta.url));
 const templatePath = path.join(__dirname, "..", "src", "agents", "agent-instructions.md");
 const template = fs.readFileSync(templatePath, "utf-8");
 
+/** Mock tools with a known, stable shape for testing */
+const mockTools: ToolDefinition[] = [
+  {
+    name: "mock-action",
+    description: [
+      "Perform a mock action.",
+      'Response: `{"ok": true}` on success.',
+    ],
+    inputSchema: {
+      type: "object",
+      properties: {
+        title: { type: "string", description: "Action title" },
+        detail: { type: "string", description: "Optional detail" },
+      },
+      required: ["title"],
+    },
+    handler: async () => ({ ok: true }),
+  },
+  {
+    name: "mock-query",
+    description: [
+      "Query mock data from the device.",
+      "Blocks until the device responds.",
+      'Response: `{"data": ...}` on success.',
+    ],
+    inputSchema: {
+      type: "object",
+      properties: {
+        tags: {
+          type: "array",
+          items: { type: "string" },
+          description: "Filter tags",
+        },
+      },
+    },
+    handler: async () => ({ data: [] }),
+  },
+];
+
 /** Minimal replica of getAgentInstructions that doesn't need host.json */
-function buildInstructions(taskId: string, skipPermissions?: boolean): string {
+function buildInstructions(taskId: string, opts?: { skipPermissions?: boolean }): string {
   let instructions = template
-    .replace(/\{\{ENDPOINT_DOCS\}\}/g, generateEndpointDocs(9966, taskId))
+    .replace(/\{\{ENDPOINT_DOCS\}\}/g, generateEndpointDocs(9966, taskId, mockTools))
     .replace(/\{\{TASK_DESCRIPTION\}\}/g, "Test task prompt");
-  if (skipPermissions) {
+  if (opts?.skipPermissions) {
     instructions = instructions.replace(/## Permissions\r?\n[\s\S]*?(?=## |\r?\n---)/m, "");
   }
   return instructions;
@@ -28,13 +67,13 @@ describe("getAgentInstructions", () => {
   });
 
   it("strips Permissions section when skipPermissions is true", () => {
-    const result = buildInstructions("test-task-id", true);
+    const result = buildInstructions("test-task-id", { skipPermissions: true });
     assert.doesNotMatch(result, /## Permissions/);
     assert.doesNotMatch(result, /PALMIER_PERMISSION/);
   });
 
   it("preserves other sections when Permissions is stripped", () => {
-    const result = buildInstructions("test-task-id", true);
+    const result = buildInstructions("test-task-id", { skipPermissions: true });
     assert.match(result, /## Reporting Output/);
     assert.match(result, /## Completion/);
     assert.match(result, /## HTTP Endpoints/);
@@ -60,22 +99,38 @@ describe("getAgentInstructions", () => {
     const result = buildInstructions("test");
     assert.match(result, /Test task prompt/);
   });
-});
 
-describe("full agent instruction snapshot", () => {
-  it("matches the expected full text exactly", () => {
-    const result = buildInstructions("test-task-id").replace(/\r\n/g, "\n").trimEnd();
-    const snapshotPath = path.join(__dirname, "fixtures", "agent-instructions-snapshot.md");
-    const expected = fs.readFileSync(snapshotPath, "utf-8").replace(/\r\n/g, "\n").trimEnd();
-    assert.equal(result, expected);
-  });
 });
 
+
 describe("generateEndpointDocs", () => {
-  const docs = generateEndpointDocs(9966, "test-id");
+  const docs = generateEndpointDocs(9966, "test-id", mockTools);
+
+  it("matches expected full output", () => {
+    const expected = [
+      "The following HTTP endpoints are available during task execution. Use curl to call them.",
+      "",
+      "**`POST http://localhost:9966/mock-action?taskId=test-id`** — Perform a mock action.",
+      "```json",
+      '{"title":"...","detail":"..."}',
+      "```",
+      "- `title` (required, string): Action title",
+      "- `detail` (optional, string): Optional detail",
+      '- Response: `{"ok": true}` on success.',
+      "",
+      "**`POST http://localhost:9966/mock-query?taskId=test-id`** — Query mock data from the device.",
+      "```json",
+      '{"tags":["..."]}',
+      "```",
+      "- `tags` (optional, string array): Filter tags",
+      "- Blocks until the device responds.",
+      '- Response: `{"data": ...}` on success.',
+    ].join("\n");
+    assert.equal(docs, expected);
+  });
 
-  it("generates docs for all MCP tools", () => {
-    for (const tool of agentTools) {
+  it("generates docs for all provided tools", () => {
+    for (const tool of mockTools) {
       assert.match(docs, new RegExp(`POST http://localhost:9966/${tool.name}\\?taskId=`), `Missing endpoint for ${tool.name}`);
     }
   });
@@ -97,8 +152,8 @@ describe("generateEndpointDocs", () => {
   });
 
   it("includes response descriptions", () => {
-    for (const tool of agentTools) {
-      if (tool.responseDescription) {
+    for (const tool of mockTools) {
+      if (tool.description.length > 1) {
         assert.match(docs, /Response:/, `Missing response description for ${tool.name}`);
       }
     }
@@ -106,6 +161,16 @@ describe("generateEndpointDocs", () => {
 
   it("marks required and optional parameters correctly", () => {
     assert.match(docs, /\(required, string\)/);
+    // "detail" has no required entry, so it should be optional
     assert.match(docs, /\(optional, string\)/);
   });
+
+  it("handles array-type parameters", () => {
+    assert.match(docs, /\(optional, string array\)/);
+    assert.match(docs, /Filter tags/);
+  });
+
+  it("renders multi-line descriptions as bullet points", () => {
+    assert.match(docs, /- Blocks until the device responds\./);
+  });
 });

package/test/agent-output-parsing.test.ts

@@ -71,3 +71,4 @@ describe("parsePermissions", () => {
   });
 });
 
+

package/test/task-parsing.test.ts

@@ -19,7 +19,7 @@ This is the task body.`;
     assert.equal(result.frontmatter.id, "abc123");
     assert.equal(result.frontmatter.name, "Test Task");
     assert.equal(result.frontmatter.agent, "claude");
-    assert.equal(result.body, "This is the task body.");
+    assert.equal(result.frontmatter.user_prompt, "Do something");
   });
 
   it("defaults agent to claude when not specified", () => {
@@ -68,7 +68,7 @@ requires_confirmation: false
     assert.throws(() => parseTaskContent("---\nname: test\n---\n"), /must include at least: id/);
   });
 
-  it("handles empty body", () => {
+  it("handles empty body gracefully", () => {
     const content = `---
 id: abc123
 user_prompt: test
@@ -78,6 +78,6 @@ requires_confirmation: false
 ---`;
 
     const result = parseTaskContent(content);
-    assert.equal(result.body, "");
+    assert.equal(result.frontmatter.id, "abc123");
   });
 });

package/dist/commands/plan-generation.md

@@ -1,22 +0,0 @@
-You are a task planning assistant. Given a task description, produce a Markdown execution plan for an AI agent to follow. Do not execute any part of the plan yourself.
-
-## Output Format
-
-Start with a YAML frontmatter block (no code fences), then the plan body:
-
----
-task_name: <concise label, 3-6 words>
----
-
-<plan body>
-
-## Plan Body Guidelines
-
-- Write a numbered sequence of concrete, actionable steps.
-- If the task produces formatted output (report, email, summary, etc.), specify the structure, sections, and tone.
-- When a step requires user input, simply state what information is needed from the user. Do **not** specify how to obtain it — the agent has its own tool for requesting user input.
-- Preserve relative time expressions (e.g., "today", "yesterday", "last week") exactly as written — do **not** resolve them to specific dates. The plan may be executed on a different day than it was generated.
-- If the task involves opening a web browser or application, include a final step to close it before finishing.
-
-## Task Description
-

package/src/commands/plan-generation.md

@@ -1,22 +0,0 @@
-You are a task planning assistant. Given a task description, produce a Markdown execution plan for an AI agent to follow. Do not execute any part of the plan yourself.
-
-## Output Format
-
-Start with a YAML frontmatter block (no code fences), then the plan body:
-
----
-task_name: <concise label, 3-6 words>
----
-
-<plan body>
-
-## Plan Body Guidelines
-
-- Write a numbered sequence of concrete, actionable steps.
-- If the task produces formatted output (report, email, summary, etc.), specify the structure, sections, and tone.
-- When a step requires user input, simply state what information is needed from the user. Do **not** specify how to obtain it — the agent has its own tool for requesting user input.
-- Preserve relative time expressions (e.g., "today", "yesterday", "last week") exactly as written — do **not** resolve them to specific dates. The plan may be executed on a different day than it was generated.
-- If the task involves opening a web browser or application, include a final step to close it before finishing.
-
-## Task Description
-

package/test/fixtures/agent-instructions-snapshot.md

@@ -1,58 +0,0 @@
-You are an AI agent executing a task on behalf of the user. Follow these instructions carefully.
-
-## Reporting Output
-
-If you generate report or output files, print each file path on its own line using this exact format:
-[PALMIER_REPORT] <filename>
-
-## Completion
-
-When you are done, output exactly one of these markers as the very last line (no other text on the same line):
-[PALMIER_TASK_SUCCESS]
-[PALMIER_TASK_FAILURE]
-
-## Permissions
-
-Whenever a tool you are trying to use is denied or you lack the required permissions, print each required permission on its own line using this exact format:
-[PALMIER_PERMISSION] <tool_name> | <description>
-
-## HTTP Endpoints
-
-The following HTTP endpoints are available during task execution. Use curl to call them.
-
-**`POST http://localhost:9966/notify?taskId=test-task-id`** — Send a push notification to the user's device.
-```json
-{"title":"...","body":"..."}
-```
-- `title` (required, string): Notification title
-- `body` (required, string): Notification body
-- Response: `{"ok": true}` on success.
-
-**`POST http://localhost:9966/request-input?taskId=test-task-id`** — Request input from the user.
-```json
-{"description":"...","questions":["..."]}
-```
-- `description` (optional, string): Context or heading for the input request
-- `questions` (required, string array): Questions to present to the user
-- The request blocks until the user responds.
-- Response: `{"values": ["answer1", "answer2"]}` on success, or `{"aborted": true}` if the user declines.
-- When you need information from the user (credentials, answers to questions, preferences, clarifications, etc.), do not guess, fail, or prompt via stdout, even in a non-interactive environment — use this endpoint instead.
-
-**`POST http://localhost:9966/request-confirmation?taskId=test-task-id`** — Request confirmation from the user.
-```json
-{"description":"..."}
-```
-- `description` (required, string): What the user is confirming
-- The request blocks until the user confirms or aborts.
-- Response: `{"confirmed": true}` or `{"confirmed": false}`.
-
-**`POST http://localhost:9966/device-geolocation?taskId=test-task-id`** — Get the GPS location of the user's mobile device.
-- When you need the user's real-time location, use this endpoint.
-- Blocks until the device responds (up to 30 seconds).
-- Response: `{"latitude": ..., "longitude": ..., "accuracy": ..., "timestamp": ...}` on success, or `{"error": "..."}` on failure.
-
-The task to execute follows below:
-
----
-
-Test task prompt