pm_prompt
{"content":"{{\u003c badge content=\"Vanguard\" color=\"blue\" class=\"tier-badge\" \u003e}}\n\nThe `pm_prompt` tool manages the prompt library — the markdown instruction files that capabilities and agents reference. Prompts have a `role`, tags, and agent bindings.\n\n## Actions\n\n`create`, `get`, `list`, `list_with_content`, `update`, `delete`, `help`\n\n## Roles\n\nEvery prompt has one of five roles:\n\n| Role | Purpose |\n|------|---------|\n| `system` | Core system prompt — agent identity and top-level instructions |\n| `guide` | Workflow guide — step-by-step instructions for a task or process |\n| `checklist` | Checklist-style constraints and verification steps |\n| `policy` | Policies, rules, and constraints the agent must follow |\n| `context` | Background context, domain knowledge, or reference material |\n\n{{\u003c callout type=\"tip\" \u003e}}\n**Context-sensitive listing.** `list` returns metadata only (id, key, role, name, tags, agentCount) to keep agent context windows small. Use `get` when you need one known prompt's actual content, or `list_with_content` when you specifically need content for multiple matching prompts.\n{{\u003c /callout \u003e}}\n\n---\n\n## create\n\nCreate a new prompt.\n\n### Parameters\n\n| Field | Type | Required | Notes |\n|-------|------|----------|-------|\n| `key` | string | Yes | Unique prompt key (max 120 chars) |\n| `role` | string | Yes | `system`, `guide`, `checklist`, `policy`, or `context` |\n| `name` | string | Yes | Display name (max 200 chars) |\n| `content` | string | Yes | Prompt content (max 20,000 chars) |\n| `tags` | string[] | No | Freeform tags |\n\n### Example\n\n```text\npm_prompt {\n action: \"create\",\n key: \"reviewer.system\",\n role: \"system\",\n name: \"Reviewer System Prompt\",\n content: \"You are a code reviewer. Your job is to...\",\n tags: [\"review\", \"core\"]\n}\n```\n\n---\n\n## get\n\nFetch a prompt by ID or key. Returns full content.\n\n### Parameters\n\n| Field | Type | Required |\n|-------|------|----------|\n| `id` | string | One of |\n| `key` | string | One of |\n\n---\n\n## list\n\n**Metadata-only list.** Returns prompt records without `content` — just id, key, role, name, tags, and `agentCount` (how many agents bind to this prompt).\n\n### Parameters\n\n| Field | Type | Required | Notes |\n|-------|------|----------|-------|\n| `roles` | string[] | No | Filter by roles array |\n| `search` | string | No | Search query |\n| `limit` | number | No | Max results (1–500) |\n\nUse this for discovery — an agent can list available prompts without blowing out its context window.\n\n### Example\n\n```text\npm_prompt {\n action: \"list\",\n roles: [\"system\"],\n search: \"reviewer\",\n limit: 20\n}\n```\n\n---\n\n## list_with_content\n\nSame filters as `list`, but each returned prompt includes full `content`. Use this only when metadata-only discovery is not enough, such as scanning the bodies of several prompts at once. For one known prompt key or id, prefer `get`.\n\n### Parameters\n\nSame as `list`: `roles`, `search`, and `limit`.\n\n---\n\n## update\n\nUpdate an existing prompt.\n\n### Parameters\n\n| Field | Type | Required |\n|-------|------|----------|\n| `id` | string | Yes |\n| `key` | string | No |\n| `role` | string | No |\n| `name` | string | No |\n| `content` | string | No |\n| `tags` | string[] | No |\n\n---\n\n## delete\n\nDelete a prompt by ID.\n\n### Parameters\n\n| Field | Type | Required |\n|-------|------|----------|\n| `id` | string | Yes |\n\n---\n\n## help\n\n```text\npm_prompt { action: \"help\" }\n```\n\n`help` returns structured, read-only guidance for prompt registry calls.\n\n| Field | Contents |\n|-------|----------|\n| `actions` | Current accepted action names: `create`, `get`, `list`, `list_with_content`, `update`, `delete`, and `help` |\n| `whenToUse` | Guidance that `list` is metadata-only, `list_with_content` is content-bearing with the same filters, and `get` loads full content by id or key |\n| `examples` | Representative calls for metadata discovery, loading a known prompt by key, and creating a prompt |\n| `cautions` | Guardrails such as generated prompt ids, content omitted from `list`, and delete clearing related bindings or capability references |\n| `compatibility` | Notes that structured help is additive and existing action names and non-help result shapes are unchanged |\n\nUse the help payload to pick the smallest prompt call that fits the context budget: `list` for discovery, `get` for one known prompt, and `list_with_content` for batch content inspection.\n\n## Related\n\n- [pm_agent](../agents/) — agents bind to prompts\n- [pm_capability](../capabilities/) — capabilities link to prompt content","description":"Store and retrieve prompt templates for agents — metadata-only list and full-content get.","title":"pm_prompt"}
Vanguard
The pm_prompt tool manages the prompt library — the markdown instruction files that capabilities and agents reference. Prompts have a role, tags, and agent bindings.
Actions
create, get, list, list_with_content, update, delete, help
Roles
Every prompt has one of five roles:
| Role | Purpose |
|---|---|
system | Core system prompt — agent identity and top-level instructions |
guide | Workflow guide — step-by-step instructions for a task or process |
checklist | Checklist-style constraints and verification steps |
policy | Policies, rules, and constraints the agent must follow |
context | Background context, domain knowledge, or reference material |
Context-sensitive listing.
list returns metadata only (id, key, role, name, tags, agentCount) to keep agent context windows small. Use get when you need one known prompt’s actual content, or list_with_content when you specifically need content for multiple matching prompts.create
Create a new prompt.
Parameters
| Field | Type | Required | Notes |
|---|---|---|---|
key | string | Yes | Unique prompt key (max 120 chars) |
role | string | Yes | system, guide, checklist, policy, or context |
name | string | Yes | Display name (max 200 chars) |
content | string | Yes | Prompt content (max 20,000 chars) |
tags | string[] | No | Freeform tags |
Example
pm_prompt {
action: "create",
key: "reviewer.system",
role: "system",
name: "Reviewer System Prompt",
content: "You are a code reviewer. Your job is to...",
tags: ["review", "core"]
}get
Fetch a prompt by ID or key. Returns full content.
Parameters
| Field | Type | Required |
|---|---|---|
id | string | One of |
key | string | One of |
list
Metadata-only list. Returns prompt records without content — just id, key, role, name, tags, and agentCount (how many agents bind to this prompt).
Parameters
| Field | Type | Required | Notes |
|---|---|---|---|
roles | string[] | No | Filter by roles array |
search | string | No | Search query |
limit | number | No | Max results (1–500) |
Use this for discovery — an agent can list available prompts without blowing out its context window.
Example
pm_prompt {
action: "list",
roles: ["system"],
search: "reviewer",
limit: 20
}list_with_content
Same filters as list, but each returned prompt includes full content. Use this only when metadata-only discovery is not enough, such as scanning the bodies of several prompts at once. For one known prompt key or id, prefer get.
Parameters
Same as list: roles, search, and limit.
update
Update an existing prompt.
Parameters
| Field | Type | Required |
|---|---|---|
id | string | Yes |
key | string | No |
role | string | No |
name | string | No |
content | string | No |
tags | string[] | No |
delete
Delete a prompt by ID.
Parameters
| Field | Type | Required |
|---|---|---|
id | string | Yes |
help
pm_prompt { action: "help" }help returns structured, read-only guidance for prompt registry calls.
| Field | Contents |
|---|---|
actions | Current accepted action names: create, get, list, list_with_content, update, delete, and help |
whenToUse | Guidance that list is metadata-only, list_with_content is content-bearing with the same filters, and get loads full content by id or key |
examples | Representative calls for metadata discovery, loading a known prompt by key, and creating a prompt |
cautions | Guardrails such as generated prompt ids, content omitted from list, and delete clearing related bindings or capability references |
compatibility | Notes that structured help is additive and existing action names and non-help result shapes are unchanged |
Use the help payload to pick the smallest prompt call that fits the context budget: list for discovery, get for one known prompt, and list_with_content for batch content inspection.
Related
- pm_agent — agents bind to prompts
- pm_capability — capabilities link to prompt content