The Agent Skills Directory

[Prompt Injection] (SAFE): The instructions in SKILL.md are purely task-oriented. There are no attempts to bypass safety filters, extract system prompts, or override agent constraints.
[Data Exposure & Exfiltration] (SAFE): No sensitive file paths, environment variables, or hardcoded credentials (API keys, tokens) were detected in any of the files.
[Unverifiable Dependencies & Remote Code Execution] (SAFE): The skill does not include any package manifest files (e.g., package.json, requirements.txt) or commands to install external software. No code execution patterns like eval(), exec(), or piped shell commands were found.
[Obfuscation] (SAFE): All content is provided in clear-text Markdown. No Base64, zero-width characters, or hex-encoded strings were detected.
[Persistence & Privilege Escalation] (SAFE): There are no commands related to system persistence (cron jobs, startup scripts) or privilege escalation (sudo, chmod).
[Indirect Prompt Injection] (LOW):
Ingestion points: The workflow in SKILL.md (Step 2) instructs the agent to use WebFetch and WebSearch to ingest external scientific literature.
Boundary markers: None explicitly defined to separate external search results from instructions.
Capability inventory: The skill is limited to text generation and does not have file-write or subprocess execution capabilities.
Sanitization: No sanitization logic is present for external data. This remains a low-level risk inherent to any skill that processes untrusted web content.

hypothesis-generation