skillboss
Fail
Audited by Gen Agent Trust Hub on Apr 17, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The configuration file
config.jsoncontains a hardcoded API key with the prefixsk-, which represents a direct exposure of sensitive credentials. - [DATA_EXFILTRATION]: Documentation in
SKILL.mdandapi-integration.mdexplicitly instructs the agent to read sensitive data from the local filesystem path~/.config/skillboss/credentials.json. This pattern encourages the exposure of user credentials stored in the standard CLI configuration location. - [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection in the 'Talking Head Video' workflow.
- Ingestion points: Untrusted search results from Perplexity and Linkup APIs are ingested into the agent context in
workflows/talking-head-video/README.md. - Boundary markers: The prompt templates (e.g.,
Research: [PASTE SEARCH RESULTS HERE]) lack explicit delimiters or instructions to ignore embedded commands within the ingested data. - Capability inventory: The skill possesses extensive capabilities including network calls via
skb api call, file system writes with the-oflag, and script generation. - Sanitization: No sanitization or validation of the external content is performed before interpolation into the script generation prompts.
Recommendations
- AI detected serious security threats
Audit Metadata