Tmp.Mw5jO0UIvc

WarnAudited by ClawScan on May 12, 2026.

Overview

The skill is clearly for Gmail, but it gives an unreviewed MCP package broad authenticated Gmail powers, including sending, forwarding, auto-replying, bulk deletion, and persistent account-setting changes without clear approval limits.

Install only if you trust the npm package and are comfortable granting it Gmail access. Pin and verify the package, check OAuth scopes, and require explicit confirmation before sending, forwarding, auto-replying, deleting mail, or changing forwarding, delegate, send-as, vacation, watch, filter, or tracking settings.

Findings (5)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

The agent could act through the user's Gmail account, including reading mail, changing labels, and sending messages.

Why it was flagged

The skill expects an already-authenticated Gmail account and selects it by environment value, but the artifacts do not state OAuth scopes, permission boundaries, or credential handling.

Skill content
- [gogcli](https://github.com/steipete/gogcli) installed and authenticated ... "GOG_ACCOUNT": "you@gmail.com"
Recommendation

Use only with a Gmail account you intend to delegate, verify OAuth scopes, and prefer a dedicated or least-privilege account if possible.

What this means

A broad or mistaken request could delete many emails or send/forward/reply to recipients unintentionally.

Why it was flagged

The skill exposes destructive and external-communication tools, including query-based bulk actions and permanent deletion, without documented confirmation, dry-run, or scope limits.

Skill content
`gog_gmail_trash` | Trash by IDs or query ... `gog_gmail_batch_delete` | Permanently delete multiple messages ... `gog_gmail_drafts_send` | Send a draft ... `gog_gmail_autoreply` | Reply once to all messages matching a query
Recommendation

Require explicit user confirmation and a preview before sending, forwarding, auto-replying, trashing, archiving, or permanently deleting messages.

What this means

A mistaken action could keep forwarding mail, granting delegate access, sending vacation responses, or tracking email activity after the user expected the task to end.

Why it was flagged

These Gmail settings can persist beyond a single chat or task, such as forwarding, delegation, vacation replies, watches, and tracking, but no approval or cleanup workflow is specified.

Skill content
Use `gog_gmail_run` for advanced settings (`filters`, `delegates`, `forwarding`, `sendas`, `vacation`, `watch`) and email tracking (`track`).
Recommendation

Treat all persistent Gmail setting changes as high-risk: require explicit approval, summarize the exact setting changed, and provide a clear rollback step.

What this means

The code that handles Gmail access may differ from what was reviewed, and a package update or compromise could affect the user's email account.

Why it was flagged

The setup fetches and runs an npm MCP package with auto-confirmation and no pinned version, while the provided artifact set contains no code files for that package.

Skill content
"command": "npx", "args": ["-y", "gogcli-mcp-gmail"]
Recommendation

Pin the package version, verify the npm package matches the claimed source repository, and review the code before granting Gmail access.

What this means

Private messages and attachments may be processed by the agent and any connected MCP tooling.

Why it was flagged

Reading raw message data and attachments through an MCP server is expected for a Gmail skill, but it means sensitive email content can enter the agent/tool context.

Skill content
`gog_gmail_raw` | Raw Gmail API JSON for a message ... `gog_gmail_attachment` | Download a single attachment
Recommendation

Avoid using the skill on highly sensitive mail unless you trust the MCP server and understand the agent platform's logging and retention behavior.