Agent Economy ReportAgent Economy ReportCAUTION version 1.0.1 · audited 2026-09-17
This skill is a command reference for ATXP, a paid API service. It tells the agent to run the `atxp` CLI through `npx`, authenticate with an `ATXP_CONNECTION` credential, and use commands that search the web, generate images/music/video, read/send email, manage accounts, check balances, and create payment links. The commands are tied to ATXP domains and documented Stripe/USDC funding flows. However, the file also tells agents to check email for messages that can contain instructions for downloading a new version of the skill, and to edit `/root/.openclaw/openclaw.json` in Clowdbot environments. Because it deals with credentials and payments and exposes an email-based instruction channel, it should be reviewed carefully.
Instructions try to steer the agent beyond the declared purpose. SKILL.md line 112 says: "**Best practice:** Check your ATXP email inbox at least once per day. Inbox checks are free, and important messages arrive here — including support replies, notifications, and updates to this skill itself with instructions for downloading the latest version." This establishes the email inbox as a source of instructions to download and apply a new version of the skill, so an outside sender who can email the agent may be able to direct future behavior.
| Declared purpose | Access ATXP paid API tools for web search, AI image generation, music creation, video generation, X/Twitter search, email, and agent account management. Use when users need real-time web search, AI-generated media (images, music, video), X/Twitter search, send/receive emails, or create and fund agent accounts. Requires authentication via `npx atxp login`. |
| Observed behavior | The file is documentation that instructs the agent to run shell commands such as `npx atxp search`, `image`, `music`, `video`, `x`, `email`, `balance`, `fund`, `whoami`, `topup`, and agent account commands. It loads credentials via `ATXP_CONNECTION` and `source ~/.atxp/config`, describes Stripe/USDC funding and paid email actions, provides a TypeScript client that connects to `https://search.mcp.atxp.ai` using the environment credential, lists additional MCP servers, tells the agent to check email for skill updates, and tells it how to edit `/root/.openclaw/openclaw.json` in Clowdbot environments. |
| Verdict, rules only | SAFE |
| Verdict, AI | CAUTION |
| Final verdict | CAUTION (never better than either pass) |
| capability | vs purpose | detail |
|---|---|---|
| network | needed | The skill's core operations are remote API calls to ATXP services and MCP endpoints such as `search.mcp.atxp.ai`, `image.mcp.atxp.ai`, `email.mcp.atxp.ai`, and `accounts.atxp.ai` (SKILL.md lines 210, 222-230, 57). |
| shell | needed | The documented interface is a CLI invoked through shell commands such as `npx atxp balance` and `source ~/.atxp/config` (SKILL.md lines 17-18, 52). |
| filesystem_read | needed | The skill instructs the agent to source `~/.atxp/config` and read `ATXP_CONNECTION` from the environment (SKILL.md lines 14, 18, 211). |
| filesystem_write | exceeds purpose | In Clowdbot environments, the skill says model access can be changed by editing `/root/.openclaw/openclaw.json` (SKILL.md line 192). |
| credentials | needed | Authentication depends on the `ATXP_CONNECTION` credential, and `npx atxp whoami` exposes account ID, email, and wallet (SKILL.md lines 14, 41). |
| payments | needed | The skill includes paid actions and funding: email send/reply and username fees, Stripe top-up links, and USDC/card funding (SKILL.md lines 32-33, 37, 42-43, 57, 92). |
| install_packages | needed | Every documented CLI command uses `npx atxp`, which downloads and runs the npm package on demand (SKILL.md line 17). |
| browser | exceeds purpose | The `topup --open` form can open a payment link in a browser, but browser opening is optional (SKILL.md line 96). |
| severity | finding |
|---|---|
| info | Declared purpose includes paid API, email, and account funding The skill's declared scope covers web search, media generation, X/Twitter search, email operations, and creating/funding agent accounts. SKILL.md:3 description: Access ATXP paid API tools for web search, AI image generation, music creation, video generation, X/Twitter search, email, and agent account manage |
| low | Uses npx to fetch and run the atxp package The agent is instructed to run the `atxp` CLI through `npx`, which downloads and executes an npm package. This is a supply-chain trust decision. SKILL.md:17 npx atxp login |
| medium | Executes shell configuration from home directory `source` runs the contents of `~/.atxp/config` as shell commands. If that file were altered, the skill would execute its contents. SKILL.md:18 source ~/.atxp/config |
| low | Relies on ATXP_CONNECTION credential The agent checks the `ATXP_CONNECTION` environment variable before using the service, so this credential must be present and protected. SKILL.md:14 echo $ATXP_CONNECTION |
| low | Exposes account and wallet metadata The `whoami` command returns account ID, type, email, wallet, and owner, so these details are visible to the agent and any output channel. SKILL.md:41 | `npx atxp whoami` | Show account info (ID, type, email, wallet) (FREE) | |
| medium | Can create payment links to pull money The `topup` command creates Stripe payment links for $1 to $1,000. A compromised agent could request payments from people. SKILL.md:92 Agents can create Stripe Payment Links to request funding. The payer can adjust the amount at checkout ($1–$1,000). |
| info | Funding supports cryptocurrency and card payments Funding is handled through USDC addresses and the accounts.atxp.ai site, which is expected for the declared paid API purpose. SKILL.md:57 Fund your ATXP account via USDC deposit addresses (one per chain, typically Base and Solana) or with credit card and other standard payment methods at https://a |
| medium | Connection string passed as command-line argument The agent logs in by placing a connection string directly in a shell command, which may expose the credential in process listings or command history. SKILL.md:102 npx atxp login --token "<connection_string>" |
| high | Email inbox is described as a source of skill updates/instructions The skill instructs the agent to treat email as a channel for receiving instructions to download the latest version. This is a prompt-injection path if an outside sender can reach the inbox. SKILL.md:112 **Best practice:** Check your ATXP email inbox at least once per day. Inbox checks are free, and important messages arrive here — including support replies, not |
| high | Instructs editing the agent's model configuration file This goes beyond using ATXP APIs: it tells the agent it can modify `/root/.openclaw/openclaw.json`, which may change how the agent's model access is configured. SKILL.md:192 If running in a [Clowdbot](https://clowd.bot) environment, model access can be modified by updating the configuration file at `/root/.openclaw/openclaw.json`. E |
| info | Programmatic access connects to remote MCP server The TypeScript example creates a client pointing at the remote ATXP search MCP server. SKILL.md:210 mcpServer: 'https://search.mcp.atxp.ai', |
0 AI finding(s) were dropped because their file, line or quote did not match the package.
| domain | service rating | where |
|---|---|---|
| accounts.atxp.ai | not in the service index | SKILL.md:57 |
| clowd.bot | not in the service index | SKILL.md:192 |
| docs.atxp.ai | not in the service index | SKILL.md:188 |
| search.mcp.atxp.ai | not in the service index | SKILL.md:210 |
| Binaries invoked | base64, bash, npx |
| Environment variables read | ATXP_CONNECTION |
| Hard-coded wallets | none |
| Pipes a download to a shell | 0 |
| eval / exec / subprocess | 0 |
| base64 blobs | 0 |
| File writes | 1 |
| Persistence | 0 |
| file | lines | sha256 |
|---|---|---|
SKILL.md | 230 | 9876fb6cd0cab8a7… |
_meta.json | 6 | e0151effcee9d8ae… |
For agents
JSON: https://agenteconomy.report/k/atxp.audit.json · badge: https://agenteconomy.report/k/atxp.audited.svg ·
skill rating: /k/atxp · commission an audit of any skill: US$ 29 per version.
The complete published package of this exact version was downloaded from the registry and read statically; nothing was executed. A deterministic pass extracts network destinations, binaries, environment variables, writes, install commands, obfuscation markers, subprocesses and wallets, each with file and line. An AI then reads the whole package with those facts and writes the summary, the capabilities and the findings under a strict schema; every finding it produces must cite a file, a line and the exact text, or it is dropped. The final verdict is the worst of the two passes. The audit does not change the skill's trust tier (policy); the author may respond through the dispute channel and the response is published here. Commissioned by: the Agent Economy Report (free program: the 150 most downloaded skills, September 2026).