Email DraftsCAUTION
A framework for building agentic apps
Overview
A framework for building agentic apps
5977febf67f1OBSERVED · 2026-10-14What it tells the agent
The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.
---
name: email-drafts
description: >-
Create, edit, and send email drafts through compose-{id} application state,
manage-draft, and send-email. Use when composing, replying, forwarding,
attaching files, applying signatures/writing style, choosing Send + Mark Done,
or checking open/click tracking on sent mail.
---
# Email Drafts
Create, edit, and manage email drafts. Each draft is stored as an application state entry keyed `compose-{id}`. The UI refreshes through the framework polling/query invalidation path and updates the compose panel automatically.
## Storage
Drafts are stored in the `application_state` SQL table via `writeAppState("compose-{id}", draft)` from `@agent-native/core/application-state`. Each entry is one draft. Multiple drafts can exist simultaneously — they appear as tabs in the compose panel.
## Calling `manage-draft`
`manage-draft`'s call arguments are NOT the stored draft record below — every
call requires a top-level `action`, and `action` decides which other fields
apply:
| `action` | Other required fields | Notes |
| -------------- | ---------------------- | ---------------------------------------- |
| `create` | none | `id` is optional; the action assigns one when omitted. Recipient/subject/body fields are all optional but should be set to compose real content. |
| `update` | `id` | `id` must be the id returned by a prior `create` call — never invent one. |
| `delete` | `id` | Removes the compose state and its saved mailbox copy. |
| `delete-saved` | `savedDraftId` | Use when only the saved mailbox copy remains (compose state already closed). |
| `delete-all` | none | Deletes every open compose draft. |
There is no standalone "create draft" tool — `create` is this action's only
entry point for a new draft. Always call `action: "create"` first and use the
`id` it returns for any later `update`/`delete` on that same draft; do not
call `update`/`delete` speculatively before a matching `create`.
To draft a reply to a specific message, create with `mode: "reply"` and
`replyToId` set to that message's id:
```json
{
"action": "create",
"to": "[email protected]",
"subject": "Re: Meeting follow-up",
"body": "Hi team,\n\nThanks for the great discussion today...",
"mode": "reply",
"replyToId": "18d4a2f9e1b2c3d4"
}
```
## Stored Draft Record (`compose-{id}`)
This is the persisted shape of the `compose-{id}` application-state row that
`manage-draft` reads and writes — not the `manage-draft` call arguments above.
```json
{
"id": "abc123",
"to": "[email protected]",
"cc": "",
"bcc": "",
"subject": "Meeting follow-up",
"body": "Hi team,\n\nThanks for the great discussion today...",
"mode": "compose",
"replyToId": "",
"replyToThreadId": ""
}
```
### Fields
| Field | Type | Required | Description |
| ----------------- | ------ | -------- | ----------------------------------------------- |
| `id` | string | yes | Unique draft ID (must match key suffix) |
| `to` | string | no | Comma-separated recipient email addresses |
| `cc` | string | no | Comma-separated CC addresses |
| `bcc` | string | no | Comma-separated BCC addresses |
| `subject` | string | no | Email subject line |
| `body` | string | no | Email body in **markdown** (see formatting below) |
| `mode` | string | no | One of: `"compose"`, `"reply"`, `"forward"` (defaults to `"compose"`) |
| `replyToId` | string | no | Message ID being replied to (for reply/forward) |
| `replyToThreadId` | string | no | Thread ID for grouping (for reply/forward) |
## Body Formatting
The `body` field uses **markdown**. The compose editor (TipTap) renders it as rich text, and the send flow converts markdown to HTML before sending via Gmail. Use standard markdown syntax:
- **Links:** `[click here](https://example.com)` — renders as a clickable hyperlink in the sent email
- **Bold:** `**bold text**`
- **Italic:** `*italic text*`
- **Lists:** `- item` (unordered) or `1. item` (ordered)
- **Headings:** `# Heading` (h1–h3)
- **Code:** `` `inline code` `` or fenced code blocks
- **Blockquotes:** `> quoted text`
- **Bare URLs:** `https://example.com` auto-links
Do NOT use raw HTML tags — use markdown only.
## Signature and Style Settings
Before creating or rewriting a draft, read the user's drafting settings with `pnpm action get-mail-settings`.
- Use `signature` exactly when it is configured; do not rewrite or duplicate it.
- If no signature is configured, omit the signature. Never derive one from the user's name, email address, or connected profile.
- Follow `writingStyle` when present.
- Keep generated copy natural and specific. Avoid generic AI email tropes, headings, and over-formal filler unless the user asks for that style.
## Updating Durable Drafting Settings
Treat requests to permanently change, add to, or remove a writing-style rule as
settings changes, not drafting requests. Route them to `update-mail-settings`.
Do not call `manage-draft`, open the compose UI, or create a draft unless the user separately asks
to create or edit an email draft.
For a durable settings request:
1. Read the current settings with `get-mail-settings`.
2. Merge the requested rule into the existing `writingStyle`, preserving
unrelated instructions. For a removal, remove only the requested rule.
3. If the requested change is ambiguous, conflicts with existing guidance, or
would materially rewrite the style, show the proposed merged wording and
ask the user to confirm before changing it.
4. Call `update-mail-settings` with the complete merged `writingStyle` value.
5. Re-reaTrust audit
CAUTIONgrade B · trust 89/100 Install with care. The audit found things worth knowing before you trust its output.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | WARN |
| L1 | Static analysis of the code | NA |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- none-observed
- Network
- none-observed
- Shell
- none-observed
- Dependencies
- pinned
- Secrets in source
- none-found
Findings (5)
.agents/skills/an
.agents/skills/content-product-development
.agents/skills/design-exploration
.agents/skills/turn-into-app
.agents/skills/visual-edit
Gates applied: no_behavioural_pass.
5977febf67f1full audit observations/trust-audit/skill/builderio__email-drafts.json · Report an issue / request a re-scanAudit history
Every audit this skill has had.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-14 | 5977febf67f1 | CAUTION | B | 89 | first audit |
Questions
What does the Email Drafts skill do?
A framework for building agentic apps
Is Email Drafts safe to install?
With care. The audit graded it B (89/100) and found 5 things worth knowing before you trust this skill, listed below with the exact line each was found on.
What can Email Drafts access on my machine?
The audit observed no filesystem, network or shell use at all in its source.
What do I need installed to use Email Drafts?
Its own instructions reference action. Dependencies are pinned to exact versions.
How current is this page?
The grade is for one exact copy of the source (5977febf67f1), read on 2026-10-14. The repository is watched, and a new audit runs when it changes — this is the first audit.
Add this badge to your README
Shows this page's audit result and links back to it. It updates when the audit does.
[](https://modelsatlas.com/skills/builderio/email-drafts/)
Not your listing, or something out of date? Claim or update your listing.