The Agent Skills Directory

[Prompt Injection] (SAFE): No malicious override instructions, bypass markers, or role-play jailbreak attempts were detected in the skill instructions.
[Data Exposure & Exfiltration] (SAFE): The skill contains explicit privacy safeguards: 'Avoid collecting/storing unnecessary personal data; never request full NI number or UTR.' No hardcoded credentials or sensitive local file path access was found.
[Indirect Prompt Injection] (LOW): The skill relies on web search results to verify volatile tax data, creating a potential surface for instructions embedded in external web content.
Ingestion points: WebSearch tool used for fetching data from gov.uk and HMRC sites.
Boundary markers: Absent; the skill does not explicitly instruct the agent to ignore instructions found within retrieved web content.
Capability inventory: Primarily text generation and mathematical calculations. No evidence of high-risk capabilities like file-writing or shell execution in the analyzed file.
Sanitization: None detected within the skill logic; relies on the underlying LLM's tool-handling guardrails.
[Unverifiable Dependencies & Remote Code Execution] (SAFE): No external package installations, remote script execution patterns, or dynamic code execution (eval/exec) were detected.
[Privilege Escalation] (SAFE): No commands related to privilege escalation (sudo, chmod) were found.

project-qtax