writing-clone-starter
Fail
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: CRITICALDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONOBFUSCATIONNO_CODE
Full Analysis
- [DATA_EXFILTRATION]: A URL referenced in
references/built-in-profiles/zhouzuoluo/raw-corpus/README.md(https://wxredian.com/art?id=07bf527cfa52a366210c61375cc30b6f) was flagged as a phishing site by automated security scanners, posing a risk of data or credential theft. - [NO_CODE]: The file
SKILL.mdwas identified as malicious by reputation scanners (FileRepMalware). While the file contains AI instructions and no executable scripts, the confirmed security detection triggers a critical verdict. - [OBFUSCATION]: Destination URLs in
references/built-in-profiles/zhouzuoluo/held-out-set.mdare hidden using Base64 encoding within the parameters of a redirect service (reading.quanlitu.com). This technique can be used to bypass static analysis of link destinations. - [INDIRECT_PROMPT_INJECTION]: The skill architecture ingests data from a 'Material Library' (
02_素材库/writing-clone-starter-material-library/), presenting a vulnerability to indirect prompt injection if malicious instructions are present in the processed files. - Ingestion points: External content is loaded from the specified material library directory into the agent context as described in
README.mdandSKILL.md. - Boundary markers: While instructions exist to separate generated text into
response.mdand summaries intorun_summary.md, there are no explicit delimiters used to encapsulate untrusted material from the library. - Capability inventory: The agent has the capability to write generated articles to the file system as defined in the
response.mdoutput contract. - Sanitization: The skill lacks explicit sanitization or filtering protocols for content ingested from the external library.
Recommendations
- CRITICAL: 1 file(s) identified as malware by FileRep - DO NOT USE
- AI detected serious security threats
- Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata