@j-o-r/hello-dave 0.0.2 → 0.0.4

package/docs.bak.1774780058/toolset.md.bak

@@ -0,0 +1,94 @@
+# ToolSet (`lib/ToolSet.js`)
+
+## Overview
+**ToolSet** manages LLM function calls (tools): registration, execution, listing. Integrates with `Prompt.js` (via `execute(prompt)`) and `AgentManager.js` (pre-configured modes, generics). Supports JSON Schema params, async methods. Exports from `lib/genericToolset.js`: `toolsPool` (9 pre-built tools).
+
+**Uses named exports from `lib/index.js` - NO default export.**
+
+Key Features:
+- **Modes**: `toolChoice: 'auto'|'none'|'required'` (default `'auto'`).
+- **Validation**: Name regex `/^[#!a-z_0-9]{2,}$/`.
+- **Events** (via `Prompt`): `tool_request`, `tool_error`, `tool_response`.
+- **Execution**: `execute(prompt)` processes `function_request`s from last message → calls → adds `function_response`s.
+
+## Constructor
+```javascript
+import { ToolSet } from '@j-o-r/hello-dave';
+
+new ToolSet(choice = 'auto');  // 'auto'|'none'|'required'
+```
+
+## Methods
+| Method | Args | Returns | Description |
+|--------|------|---------|-------------|
+| `add(name, desc, params, method)` | `string`, `string`, `TSSchema`, `async (params) => *` | `void` | Register tool. Overwrites if exists. |
+| `get(name)` | `string` | `TSTool` | Get tool details. |
+| `delete(name)` | `string` | `void` | Remove tool. |
+| `has(name)` | `string` | `boolean` | Exists? |
+| `list()` | - | `TSToolListItem[]` (sorted) | All tools (name/desc/params). |
+| `call(name, params)` | `string`, `object` | `Promise<*>` | Execute tool. |
+| `execute(prompt)` | `Prompt` | `Promise<void>` | Process last message `function_request`s → execute → add `function_response`s to `tool` role. |
+| `length` | - | `number` | Tool count. |
+| `toolChoice` | - | `string` | Getter for mode. |
+
+**TSTool**: `{description: string, parameters: TSSchema, method: async (params) => *}`  
+**TSSchema**: OpenAI-style JSON Schema (`{type: 'object', properties: {...}, required: [...]}`).
+
+## Generic Tools (`lib/genericToolset.js` → `toolsPool`)
+Pre-built `ToolSet('auto')` with 9 tools. Copied via `AgentManager.addGenericToolcall(name)`.
+
+| Tool Name                | Description |
+|--------------------------|-------------|
+| `javascript_interpreter` | Execute ESM ES6 JavaScript on `node`. `console.log` captures output. cwd: `~/devpri/js/hello-dave`. |
+| `get_user_env`           | Get user environment (name, system, city/region/country/timezone, external IP). |
+| `execute_bash_script`    | Execute raw Bash script/command (no escaping; verbatim `$ | < > & " ' \\ \` newlines $(( )) [[ ]]`. Heredoc-safe). Ubuntu 25.10. |
+| `send_email`             | Send email via `msmtp` (to, subject, body). |
+| `open_link`              | Open URL/file with `xdg-open`. |
+| `execute_remote_script`  | Execute raw Bash on remote via SSH (`ssh://user@host[:port]`). |
+| `history_search`         | Search chat sessions: `"(todo\|task)"` or `"package.json"`. Regex in `.cache/[app]/[prompt]/sessions/*.ndjson`. |
+| `read_file`              | Read raw file in CWD (relative path, no `/ .. \\\\`). |
+| `write_file`             | Write raw content to file in CWD (relative, as-is; no escaping). Returns bytes written. |
+
+## Integration with AgentManager
+In `setup()`:
+```javascript
+agent.setup({
+  toolsetMode: 'auto',  // Loads genericToolset.js equivalents
+  // or 'required': Empty ToolSet (add manually)
+});
+agent.addGenericToolcall('read_file');  // Copy from toolsPool post-setup
+```
+
+**Custom Tools**:
+```javascript
+import { AgentManager } from '@j-o-r/hello-dave';
+
+agent.setup({ toolsetMode: 'required' });
+const ts = agent.getToolset();
+ts.add('math', 'Add numbers', {
+  type: 'object',
+  properties: {a: {type: 'number'}, b: {type: 'number'}},
+  required: ['a', 'b']
+}, async ({a, b}) => a + b);
+```
+
+**Workflow** (in `Prompt.call()` / API adapters):
+1. LLM → `function_request`s (assistant message).
+2. `ToolSet.execute(prompt)` → parallel `call()` → `function_response`s (tool message).
+3. Recurse until text response.
+
+## Examples
+### Standalone ToolSet
+```javascript
+import { ToolSet } from '@j-o-r/hello-dave';
+
+const ts = new ToolSet('required');
+ts.add('hello', 'Say hello', {type: 'object', properties: {name: {type: 'string'}}, required: ['name']},
+  async ({name}) => `Hello, ${name}!`);
+console.log(await ts.call('hello', {name: 'World'}));  // "Hello, World!"
+```
+
+### With Prompt/AgentManager
+See [agent-manager.md](../agent-manager.md#custom-tools--generics), [prompt-class.md](../prompt-class.md).
+
+**Updated:** March 24, 2026 (9 generics from `lib/genericToolset.js`).

package/docs.bak.1774780058/toolset.md.bak3

@@ -0,0 +1,161 @@
+# ToolSet (\`lib/ToolSet.js\`)
+
+## Overview
+**ToolSet** manages LLM function calls (tools): registration, execution, listing. Integrates with \`Prompt.js\` (via \`execute(prompt)\`) and \`AgentManager.js\` (pre-configured modes, generics). Supports JSON Schema params, async methods. Exports from \`lib/genericToolset.js\`: \`toolsPool\` (11 pre-built tools).
+
+**Uses named exports from \`lib/index.js\` - NO default export.**
+
+Key Features:
+- **Modes**: \`toolChoice: 'auto'|'none'|'required'\` (default \`'auto'\`).
+- **Validation**: Name regex \`/^\[#![a-z_0-9]{2,}$/\`.
+- **Events** (via \`Prompt\`): \`tool_request\`, \`tool_error\`, \`tool_response\`.
+- **Execution**: \`execute(prompt)\` processes \`function_request\`s from last message → calls → adds \`function_response\`s.
+
+## Constructor
+\`\`\`javascript
+import { ToolSet } from '@j-o-r/hello-dave';
+
+new ToolSet(choice = 'auto');  // 'auto'|'none'|'required'
+\`\`\`
+
+## Methods
+| Method | Args | Returns | Description |
+|--------|------|---------|-------------|
+| \`add(name, desc, params, method)\` | \`string\`, \`string\`, \`TSSchema\`, \`async (params) => *\` | \`void\` | Register tool. Overwrites if exists. |
+| \`get(name)\` | \`string\` | \`TSTool\` | Get tool details. |
+| \`delete(name)\` | \`string\` | \`void\` | Remove tool. |
+| \`has(name)\` | \`string\` | \`boolean\` | Exists? |
+| \`list()\` | - | \`TSToolListItem[]\` (sorted) | All tools (name/desc/params). |
+| \`call(name, params)\` | \`string\`, \`object\` | \`Promise<*\>\` | Execute tool. |
+| \`execute(prompt)\` | \`Prompt\` | \`Promise<void>\` | Process last message \`function_request\`s → execute → add \`function_response\`s to \`tool\` role. |
+| \`length\` | - | \`number\` | Tool count. |
+| \`toolChoice\` | - | \`string\` | Getter for mode. |
+
+**TSTool**: \`{description: string, parameters: TSSchema, method: async (params) => *}\`  
+**TSSchema**: OpenAI-style JSON Schema (\`{type: 'object', properties: {...}, required: [...]}\`).
+
+## Generic Tools (\`lib/genericToolset.js\` → \`toolsPool\`)
+Pre-built \`ToolSet('auto')\` with **11 tools**. Copied via \`AgentManager.addGenericToolcall(name)\` or \`toolsetMode: 'auto'\`.
+
+| Tool Name                  | Description |
+|----------------------------|-------------|
+| \`javascript_interpreter\` | Execute ESM ES6 JavaScript on \`node\`. \`console.log\` captures output. cwd: \`~/devpri/js/hello-dave\`. |
+| \`get_user_env\`           | Get user environment (name, system, city/region/country/timezone, external IP). |
+| \`execute_bash_script\`    | Execute raw Bash script/command (no escaping; verbatim \`$\| < > & " ' \\\\` newlines \`$(())\` \`[[ ]]\`. Heredoc-safe). Ubuntu 25.10. |
+| \`send_email\`             | Send email via \`msmtp\` (to, subject, body). |
+| \`open_link\`              | Open URL/file with \`xdg-open\`. |
+| \`execute_remote_script\`  | Execute raw Bash on remote via SSH (\`ssh://user@host[:port]\`). |
+| \`history_search\`         | Search chat sessions: \`"(todo\\|task)"\` or \`"package.json"\`. Regex in \`.cache/[app]/[prompt]/sessions/*.ndjson\`. |
+| \`read_file\`              | Read raw file in CWD (relative path, no \`/\` \`..\` \`\\\\\`). |
+| \`write_file\`             | Write raw content to file in CWD (relative, as-is; no escaping). Returns bytes written. |
+| \`memory_recall\`          | Recall persistent memory entries from agent-specific storage (\`.cache/[agent]/memory.ndjson\`). |
+| \`memory_write\`           | Write/update persistent memory entries for long-term state (\`.cache/[agent]/memory.ndjson\`). |
+
+## Memory Tools (New)
+These tools provide persistent, agent-specific memory storage in NDJSON format (\`.cache/[agent-name]/memory.ndjson\`). Append-only; latest value per key. Ideal for token efficiency by offloading state/decisions from context.
+
+### \`memory_recall\`
+**Description**: Retrieve the latest values for specified keys from memory file. Returns JSON: \`{key: value, ...}\` or \`{}\` if missing.
+
+**Parameters**:
+\`\`\`json
+{
+  "type": "object",
+  "properties": {
+    "keys": {
+      "type": "array",
+      "items": { "type": "string" },
+      "description": "List of memory keys to recall, e.g., ['current_task', 'last_decision']"
+    }
+  },
+  "required": ["keys"]
+}
+\`\`\`
+
+**Example Usage in Agent Prompt**:
+\`\`\`
+Before proceeding, recall your current task and previous decision:
+- Call memory_recall with keys: ["current_task", "last_decision"]
+Then, use the recalled info to avoid repetition.
+\`\`\`
+
+**Token Efficiency**: Stores tasks/decisions externally. Recall only needed state (e.g., 100 tokens) vs. full history (10k+ tokens), preventing loops/re-reasoning.
+
+### \`memory_write\`
+**Description**: Append/update key-value pairs to memory (overwrites prior same-key entries). Returns confirmation with stored entries.
+
+**Parameters**:
+\`\`\`json
+{
+  "type": "object",
+  "properties": {
+    "entries": {
+      "type": "array",
+      "items": {
+        "type": "object",
+        "properties": {
+          "key": { "type": "string", "description": "Unique key (e.g., 'current_task')" },
+          "value": { "type": "string", "description": "JSON-stringifiable value" }
+        },
+        "required": ["key", "value"]
+      },
+      "description": "List of {key, value} pairs to store"
+    }
+  },
+  "required": ["entries"]
+}
+\`\`\`
+
+**Example Usage in Agent Prompt**:
+\`\`\`
+After analyzing, store the plan:
+- Call memory_write with entries: [{"key": "current_task", "value": "Implement login feature"}, {"key": "step", "value": "1/5"}]
+\`\`\`
+
+**Token Efficiency**: Persist intermediate results across calls/sessions. Avoids context explosion/loops by recalling compact state instead of regenerating.
+
+## Integration with AgentManager
+In \`setup()\`:
+\`\`\`javascript
+agent.setup({
+  toolsetMode: 'auto',  // Loads all 11 generic tools
+  // or 'required': Empty ToolSet (add manually)
+});
+agent.addGenericToolcall('memory_recall');  // Selective copy post-setup
+\`\`\`
+
+**Custom Tools**:
+\`\`\`javascript
+import { AgentManager } from '@j-o-r/hello-dave';
+
+agent.setup({ toolsetMode: 'required' });
+const ts = agent.getToolset();
+ts.add('math', 'Add numbers', {
+  type: 'object',
+  properties: {a: {type: 'number'}, b: {type: 'number'}},
+  required: ['a', 'b']
+}, async ({a, b}) => a + b);
+\`\`\`
+
+**Workflow** (in \`Prompt.call()\` / API adapters):
+1. LLM → \`function_request\`s (assistant message).
+2. \`ToolSet.execute(prompt)\` → parallel \`call()\` → \`function_response\`s (tool message).
+3. Recurse until text response.
+
+## Examples
+### Standalone ToolSet
+\`\`\`javascript
+import { ToolSet } from '@j-o-r/hello-dave';
+
+const ts = new ToolSet('required');
+ts.add('hello', 'Say hello', {type: 'object', properties: {name: {type: 'string'}}, required: ['name']},
+  async ({name}) => \`Hello, \${name}!\`);
+console.log(await ts.call('hello', {name: 'World'}));  // "Hello, World!"
+\`\`\`
+
+### With Prompt/AgentManager
+See [AgentManager](./agent-manager.md#custom-tools--generics), [Prompt](./prompt-class.md).
+
+**Updated:** March 26, 2026 (11 generics from \`lib/genericToolset.js\`).
+
+See also: [GenericToolset Source](./generic-toolset.md).

package/docs.bak.1774780058/toolset.md.bak4

@@ -0,0 +1,161 @@
+# ToolSet (\`lib/ToolSet.js\`)
+
+## Overview
+**ToolSet** manages LLM function calls (tools): registration, execution, listing. Integrates with \`Prompt.js\` (via \`execute(prompt)\`) and \`AgentManager.js\` (pre-configured modes, generics). Supports JSON Schema params, async methods. Exports from \`lib/genericToolset.js\`: \`toolsPool\` (11 pre-built tools).
+
+**Uses named exports from \`lib/index.js\` - NO default export.**
+
+Key Features:
+- **Modes**: \`toolChoice: 'auto'|'none'|'required'\` (default \`'auto'\`).
+- **Validation**: Name regex \`/^\[#![a-z_0-9]{2,}$/\`.
+- **Events** (via \`Prompt\`): \`tool_request\`, \`tool_error\`, \`tool_response\`.
+- **Execution**: \`execute(prompt)\` processes \`function_request\`s from last message → calls → adds \`function_response\`s.
+
+## Constructor
+\`\`\`javascript
+import { ToolSet } from '@j-o-r/hello-dave';
+
+new ToolSet(choice = 'auto');  // 'auto'|'none'|'required'
+\`\`\`
+
+## Methods
+| Method | Args | Returns | Description |
+|--------|------|---------|-------------|
+| \`add(name, desc, params, method)\` | \`string\`, \`string\`, \`TSSchema\`, \`async (params) => *\` | \`void\` | Register tool. Overwrites if exists. |
+| \`get(name)\` | \`string\` | \`TSTool\` | Get tool details. |
+| \`delete(name)\` | \`string\` | \`void\` | Remove tool. |
+| \`has(name)\` | \`string\` | \`boolean\` | Exists? |
+| \`list()\` | - | \`TSToolListItem[]\` (sorted) | All tools (name/desc/params). |
+| \`call(name, params)\` | \`string\`, \`object\` | \`Promise<*\>\` | Execute tool. |
+| \`execute(prompt)\` | \`Prompt\` | \`Promise<void>\` | Process last message \`function_request\`s → execute → add \`function_response\`s to \`tool\` role. |
+| \`length\` | - | \`number\` | Tool count. |
+| \`toolChoice\` | - | \`string\` | Getter for mode. |
+
+**TSTool**: \`{description: string, parameters: TSSchema, method: async (params) => *}\`  
+**TSSchema**: OpenAI-style JSON Schema (\`{type: 'object', properties: {...}, required: [...]}\`).
+
+## Generic Tools (\`lib/genericToolset.js\` → \`toolsPool\`)
+Pre-built \`ToolSet('auto')\` with **11 tools**. Copied via \`AgentManager.addGenericToolcall(name)\` or \`toolsetMode: 'auto'\`.
+
+| Tool Name                  | Description |
+|----------------------------|-------------|
+| \`javascript_interpreter\` | Execute ESM ES6 JavaScript on \`node\`. \`console.log\` captures output. cwd: \`~/devpri/js/hello-dave\`. |
+| \`get_user_env\`           | Get user environment (name, system, city/region/country/timezone, external IP). |
+| \`execute_bash_script\`    | Execute raw Bash script/command (no escaping; verbatim \`$\| < > & " ' \\\\` newlines \`$(())\` \`[[ ]]\`. Heredoc-safe). Ubuntu 25.10. |
+| \`send_email\`             | Send email via \`msmtp\` (to, subject, body). |
+| \`open_link\`              | Open URL/file with \`xdg-open\`. |
+| \`execute_remote_script\`  | Execute raw Bash on remote via SSH (\`ssh://user@host[:port]\`). |
+| \`history_search\`         | Search chat sessions: \`"(todo\\|task)"\` or \`"package.json"\`. Regex in \`.cache/[app]/[prompt]/sessions/*.ndjson\`. |
+| \`read_file\`              | Read raw file in CWD (relative path, no \`/\` \`..\` \`\\\\\`). |
+| \`write_file\`             | Write raw content to file in CWD (relative, as-is; no escaping). Returns bytes written. |
+| \`memory_recall\`          | Recall persistent memory entries from agent-specific storage (\`.cache/[agent]/memory.ndjson\`). |
+| \`memory_write\`           | Write/update persistent memory entries for long-term state (\`.cache/[agent]/memory.ndjson\`). |
+
+## Memory Tools (New)
+These tools provide persistent, agent-specific memory storage in NDJSON format (\`.cache/[agent-name]/memory.ndjson\`). Append-only; latest value per key. Ideal for token efficiency by offloading state/decisions from context.
+
+### \`memory_recall\`
+**Description**: Retrieve the latest values for specified keys from memory file. Returns JSON: \`{key: value, ...}\` or \`{}\` if missing.
+
+**Parameters**:
+\`\`\`json
+{
+  "type": "object",
+  "properties": {
+    "keys": {
+      "type": "array",
+      "items": { "type": "string" },
+      "description": "List of memory keys to recall, e.g., ['current_task', 'last_decision']"
+    }
+  },
+  "required": ["keys"]
+}
+\`\`\`
+
+**Example Usage in Agent Prompt**:
+\`\`\`
+Before proceeding, recall your current task and previous decision:
+- Call memory_recall with keys: ["current_task", "last_decision"]
+Then, use the recalled info to avoid repetition.
+\`\`\`
+
+**Token Efficiency**: Stores tasks/decisions externally. Recall only needed state (e.g., 100 tokens) vs. full history (10k+ tokens), preventing loops/re-reasoning.
+
+### \`memory_write\`
+**Description**: Append/update key-value pairs to memory (overwrites prior same-key entries). Returns confirmation with stored entries.
+
+**Parameters**:
+\`\`\`json
+{
+  "type": "object",
+  "properties": {
+    "entries": {
+      "type": "array",
+      "items": {
+        "type": "object",
+        "properties": {
+          "key": { "type": "string", "description": "Unique key (e.g., 'current_task')" },
+          "value": { "type": "string", "description": "JSON-stringifiable value" }
+        },
+        "required": ["key", "value"]
+      },
+      "description": "List of {key, value} pairs to store"
+    }
+  },
+  "required": ["entries"]
+}
+\`\`\`
+
+**Example Usage in Agent Prompt**:
+\`\`\`
+After analyzing, store the plan:
+- Call memory_write with entries: [{"key": "current_task", "value": "Implement login feature"}, {"key": "step", "value": "1/5"}]
+\`\`\`
+
+**Token Efficiency**: Persist intermediate results across calls/sessions. Avoids context explosion/loops by recalling compact state instead of regenerating.
+
+## Integration with AgentManager
+In \`setup()\`:
+\`\`\`javascript
+agent.setup({
+  toolsetMode: 'auto',  // Loads all 11 generic tools
+  // or 'required': Empty ToolSet (add manually)
+});
+agent.addGenericToolcall('memory_recall');  // Selective copy post-setup
+\`\`\`
+
+**Custom Tools**:
+\`\`\`javascript
+import { AgentManager } from '@j-o-r/hello-dave';
+
+agent.setup({ toolsetMode: 'required' });
+const ts = agent.getToolset();
+ts.add('math', 'Add numbers', {
+  type: 'object',
+  properties: {a: {type: 'number'}, b: {type: 'number'}},
+  required: ['a', 'b']
+}, async ({a, b}) => a + b);
+\`\`\`
+
+**Workflow** (in \`Prompt.call()\` / API adapters):
+1. LLM → \`function_request\`s (assistant message).
+2. \`ToolSet.execute(prompt)\` → parallel \`call()\` → \`function_response\`s (tool message).
+3. Recurse until text response.
+
+## Examples
+### Standalone ToolSet
+\`\`\`javascript
+import { ToolSet } from '@j-o-r/hello-dave';
+
+const ts = new ToolSet('required');
+ts.add('hello', 'Say hello', {type: 'object', properties: {name: {type: 'string'}}, required: ['name']},
+  async ({name}) => \`Hello, \${name}!\`);
+console.log(await ts.call('hello', {name: 'World'}));  // "Hello, World!"
+\`\`\`
+
+### With Prompt/AgentManager
+See [AgentManager](./agent-manager.md#custom-tools--generics), [Prompt](./prompt-class.md).
+
+**Updated:** March 26, 2026 (11 generics from \`lib/genericToolset.js\`).
+
+See also: [GenericToolset Source](./generic-toolset.md).

package/docs.bak.1774780058/toolset.md.bak5

@@ -0,0 +1,161 @@
+# ToolSet (\`lib/ToolSet.js\`)
+
+## Overview
+**ToolSet** manages LLM function calls (tools): registration, execution, listing. Integrates with \`Prompt.js\` (via \`execute(prompt)\`) and \`AgentManager.js\` (pre-configured modes, generics). Supports JSON Schema params, async methods. Exports from \`lib/genericToolset.js\`: \`toolsPool\` (11 pre-built tools).
+
+**Uses named exports from \`lib/index.js\` - NO default export.**
+
+Key Features:
+- **Modes**: \`toolChoice: 'auto'|'none'|'required'\` (default \`'auto'\`).
+- **Validation**: Name regex \`/^\[#![a-z_0-9]{2,}$/\`.
+- **Events** (via \`Prompt\`): \`tool_request\`, \`tool_error\`, \`tool_response\`.
+- **Execution**: \`execute(prompt)\` processes \`function_request\`s from last message → calls → adds \`function_response\`s.
+
+## Constructor
+\`\`\`javascript
+import { ToolSet } from '@j-o-r/hello-dave';
+
+new ToolSet(choice = 'auto');  // 'auto'|'none'|'required'
+\`\`\`
+
+## Methods
+| Method | Args | Returns | Description |
+|--------|------|---------|-------------|
+| \`add(name, desc, params, method)\` | \`string\`, \`string\`, \`TSSchema\`, \`async (params) => *\` | \`void\` | Register tool. Overwrites if exists. |
+| \`get(name)\` | \`string\` | \`TSTool\` | Get tool details. |
+| \`delete(name)\` | \`string\` | \`void\` | Remove tool. |
+| \`has(name)\` | \`string\` | \`boolean\` | Exists? |
+| \`list()\` | - | \`TSToolListItem[]\` (sorted) | All tools (name/desc/params). |
+| \`call(name, params)\` | \`string\`, \`object\` | \`Promise<*\>\` | Execute tool. |
+| \`execute(prompt)\` | \`Prompt\` | \`Promise<void>\` | Process last message \`function_request\`s → execute → add \`function_response\`s to \`tool\` role. |
+| \`length\` | - | \`number\` | Tool count. |
+| \`toolChoice\` | - | \`string\` | Getter for mode. |
+
+**TSTool**: \`{description: string, parameters: TSSchema, method: async (params) => *}\`  
+**TSSchema**: OpenAI-style JSON Schema (\`{type: 'object', properties: {...}, required: [...]}\`).
+
+## Generic Tools (\`lib/genericToolset.js\` → \`toolsPool\`)
+Pre-built \`ToolSet('auto')\` with **11 tools**. Copied via \`AgentManager.addGenericToolcall(name)\` or \`toolsetMode: 'auto'\`.
+
+| Tool Name                  | Description |
+|----------------------------|-------------|
+| \`javascript_interpreter\` | Execute ESM ES6 JavaScript on \`node\`. \`console.log\` captures output. cwd: \`~/devpri/js/hello-dave\`. |
+| \`get_user_env\`           | Get user environment (name, system, city/region/country/timezone, external IP). |
+| \`execute_bash_script\`    | Execute raw Bash script/command (no escaping; verbatim \`$\| < > & " ' \\\\` newlines \`$(())\` \`[[ ]]\`. Heredoc-safe). Ubuntu 25.10. |
+| \`send_email\`             | Send email via \`msmtp\` (to, subject, body). |
+| \`open_link\`              | Open URL/file with \`xdg-open\`. |
+| \`execute_remote_script\`  | Execute raw Bash on remote via SSH (\`ssh://user@host[:port]\`). |
+| \`history_search\`         | Search chat sessions: \`"(todo\\|task)"\` or \`"package.json"\`. Regex in \`.cache/[app]/[prompt]/sessions/*.ndjson\`. |
+| \`read_file\`              | Read raw file in CWD (relative path, no \`/\` \`..\` \`\\\\\`). |
+| \`write_file\`             | Write raw content to file in CWD (relative, as-is; no escaping). Returns bytes written. |
+| \`memory_recall\`          | Recall persistent memory entries from agent-specific storage (\`.cache/[agent]/memory.ndjson\`). |
+| \`memory_write\`           | Write/update persistent memory entries for long-term state (\`.cache/[agent]/memory.ndjson\`). |
+
+## Memory Tools (New)
+These tools provide persistent, agent-specific memory storage in NDJSON format (\`.cache/[agent-name]/memory.ndjson\`). Append-only; latest value per key. Ideal for token efficiency by offloading state/decisions from context.
+
+### \`memory_recall\`
+**Description**: Retrieve the latest values for specified keys from memory file. Returns JSON: \`{key: value, ...}\` or \`{}\` if missing.
+
+**Parameters**:
+\`\`\`json
+{
+  "type": "object",
+  "properties": {
+    "keys": {
+      "type": "array",
+      "items": { "type": "string" },
+      "description": "List of memory keys to recall, e.g., ['current_task', 'last_decision']"
+    }
+  },
+  "required": ["keys"]
+}
+\`\`\`
+
+**Example Usage in Agent Prompt**:
+\`\`\`
+Before proceeding, recall your current task and previous decision:
+- Call memory_recall with keys: ["current_task", "last_decision"]
+Then, use the recalled info to avoid repetition.
+\`\`\`
+
+**Token Efficiency**: Stores tasks/decisions externally. Recall only needed state (e.g., 100 tokens) vs. full history (10k+ tokens), preventing loops/re-reasoning.
+
+### \`memory_write\`
+**Description**: Append/update key-value pairs to memory (overwrites prior same-key entries). Returns confirmation with stored entries.
+
+**Parameters**:
+\`\`\`json
+{
+  "type": "object",
+  "properties": {
+    "entries": {
+      "type": "array",
+      "items": {
+        "type": "object",
+        "properties": {
+          "key": { "type": "string", "description": "Unique key (e.g., 'current_task')" },
+          "value": { "type": "string", "description": "JSON-stringifiable value" }
+        },
+        "required": ["key", "value"]
+      },
+      "description": "List of {key, value} pairs to store"
+    }
+  },
+  "required": ["entries"]
+}
+\`\`\`
+
+**Example Usage in Agent Prompt**:
+\`\`\`
+After analyzing, store the plan:
+- Call memory_write with entries: [{"key": "current_task", "value": "Implement login feature"}, {"key": "step", "value": "1/5"}]
+\`\`\`
+
+**Token Efficiency**: Persist intermediate results across calls/sessions. Avoids context explosion/loops by recalling compact state instead of regenerating.
+
+## Integration with AgentManager
+In \`setup()\`:
+\`\`\`javascript
+agent.setup({
+  toolsetMode: 'auto',  // Loads all 11 generic tools
+  // or 'required': Empty ToolSet (add manually)
+});
+agent.addGenericToolcall('memory_recall');  // Selective copy post-setup
+\`\`\`
+
+**Custom Tools**:
+\`\`\`javascript
+import { AgentManager } from '@j-o-r/hello-dave';
+
+agent.setup({ toolsetMode: 'required' });
+const ts = agent.getToolset();
+ts.add('math', 'Add numbers', {
+  type: 'object',
+  properties: {a: {type: 'number'}, b: {type: 'number'}},
+  required: ['a', 'b']
+}, async ({a, b}) => a + b);
+\`\`\`
+
+**Workflow** (in \`Prompt.call()\` / API adapters):
+1. LLM → \`function_request\`s (assistant message).
+2. \`ToolSet.execute(prompt)\` → parallel \`call()\` → \`function_response\`s (tool message).
+3. Recurse until text response.
+
+## Examples
+### Standalone ToolSet
+\`\`\`javascript
+import { ToolSet } from '@j-o-r/hello-dave';
+
+const ts = new ToolSet('required');
+ts.add('hello', 'Say hello', {type: 'object', properties: {name: {type: 'string'}}, required: ['name']},
+  async ({name}) => \`Hello, \${name}!\`);
+console.log(await ts.call('hello', {name: 'World'}));  // "Hello, World!"
+\`\`\`
+
+### With Prompt/AgentManager
+See [AgentManager](./agent-manager.md#custom-tools--generics), [Prompt](./prompt-class.md).
+
+**Updated:** March 26, 2026 (11 generics from \`lib/genericToolset.js\`).
+
+See also: [GenericToolset Source](./generic-toolset.md).