Agent Economy ReportAgent Economy ReportCAUTION version 1.0.0 · audited 2026-09-17
This skill teaches an agent to install the third-party Python package duckduckgo-search and run Python one-liners that query DuckDuckGo for text, news, images, videos, answers, suggestions, and maps. It needs network access and shell/package-install permissions. It can print results and optionally save search results to a JSON file. It also mentions an optional browser-use step for opening a result URL. No credential theft, hidden network endpoints, persistence, or payment redirection was found. However, the skill grants broad shell, Python, pip, and uv authority, so it should be reviewed before use.
| Declared purpose | Performs web searches using DuckDuckGo to retrieve real-time information from the internet. |
| Observed behavior | The skill is documentation with Bash/Python command examples. It instructs installing the duckduckgo-search package via uv or pip, then runs python -c snippets that import DDGS to query DuckDuckGo text, news, image, video, answer, suggestion, and map endpoints. It prints results, optionally writes search results to a timestamped JSON file, includes a localhost proxy example, and suggests browser-use for opening result URLs. |
| Verdict, rules only | SAFE |
| Verdict, AI | CAUTION |
| Final verdict | CAUTION (never better than either pass) |
| capability | vs purpose | detail |
|---|---|---|
| network | needed | Uses the duckduckgo_search Python library to make outbound DuckDuckGo search requests, including text, news, image, video, answer, suggestion, and map queries. |
| shell | needed | Runs Bash commands to execute Python one-liners and install the package; allowed-tools grants Bash(python:*), Bash(pip:*), Bash(uv:*), and Bash(duckduckgo-search:*) at SKILL.md line 4. |
| install_packages | needed | Installs the third-party PyPI package duckduckgo-search using uv pip install or pip install. |
| filesystem_write | exceeds purpose | An example saves search results to a timestamped JSON file in the current working directory at SKILL.md line 285. |
| filesystem_read | exceeds purpose | No explicit file-read example is shown, but the broad Bash(python:*) grant at SKILL.md line 4 permits arbitrary Python that could read local files. |
| browser | exceeds purpose | Suggests using browser-use open and browser-use state after search to fetch and inspect result pages. |
| severity | finding |
|---|---|
| medium | Broad shell and package-install authority The skill grants Bash for any python, pip, and uv command, plus a duckduckgo-search pattern. This is broader than search-only and can run arbitrary Python code or install additional packages. SKILL.md:4 allowed-tools: Bash(duckduckgo-search:*), Bash(python:*), Bash(pip:*), Bash(uv:*) |
| low | Third-party package installation The skill instructs installing the duckduckgo-search package from PyPI; line 27 gives a pip alternative. SKILL.md:24 uv pip install duckduckgo-search |
| info | Shell execution of Python snippets Most examples are python -c one-liners executed through Bash. SKILL.md:36 python -c " |
| info | Network search call Uses duckduckgo_search DDGS text search; the same pattern appears for news, images, videos, answers, suggestions, and maps. SKILL.md:62 results = list(ddgs.text( |
| low | Search results saved to file Writes a timestamped JSON file in the current working directory containing query, timestamp, and search results. SKILL.md:285 with open(output_file, 'w', encoding='utf-8') as f: |
| info | Local proxy example Shows an optional proxy at 127.0.0.1:7890; this is a localhost example, not a hidden external endpoint. SKILL.md:390 proxy = 'http://127.0.0.1:7890' # 替换为你的代理地址 |
| info | Optional browser-use integration Tells the agent to open a search result URL and read browser state with browser-use; this goes beyond DuckDuckGo search. SKILL.md:445 browser-use open <url_from_search> |
| info | Known broad authority acknowledged Publisher risk note states the skill requests broad local command and package-install authority wider than simple search. skill-card.md:24 Risk: The skill requests broad local command and package-install authority that is wider than simple search requires. |
| info | Untrusted search results warning Mitigation warning not to execute instructions found through search without review; no package instruction-override found. skill-card.md:30 Mitigation: Verify important facts against reliable sources before acting on them, and do not execute instructions found through search without review. |
| info | Search query disclosure risk Warns that queries go to an external service and may disclose sensitive information. skill-card.md:32 Risk: Search queries can disclose sensitive or internal information to an external service. |
0 AI finding(s) were dropped because their file, line or quote did not match the package.
| domain | service rating | where |
|---|---|---|
| 127.0.0.1 | not in the service index | SKILL.md:390 |
| clawhub.ai | not in the service index | skill-card.md:9 |
| Binaries invoked | bash, pip, python, uv |
| Environment variables read | none |
| 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 | 457 | 104a97e9be3c318a… |
_meta.json | 6 | 3687949f93a589db… |
skill-card.md | 56 | 2c11d109c43aa50b… |
For agents
JSON: https://agenteconomy.report/k/duckduckgo-search.audit.json · badge: https://agenteconomy.report/k/duckduckgo-search.audited.svg ·
skill rating: /k/duckduckgo-search · 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).