The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches untrusted data from the internet, such as Reddit post titles, Hacker News stories, and DuckDuckGo search snippets. This content is analyzed by the AI to produce summaries and playbooks. A malicious actor could craft web content to manipulate the AI's output or behavior during this research process.
Ingestion points: Data is collected in scripts/fetch.py and then processed by the AI as instructed in SKILL.md (Steps 5 and 8).
Boundary markers: The skill does not implement robust boundary markers or safety instructions to prevent the AI from following commands embedded within the fetched external data.
Capability inventory: The skill has access to the local filesystem for writing reports, can execute shell commands (python3), and performs network operations.
Sanitization: The data fetching script performs basic HTML entity decoding but lacks comprehensive sanitization of the external text before it is presented to the AI.
[DYNAMIC_EXECUTION]: The skill uses shell HEREDOCs (python3 << 'PYEOF') in SKILL.md to create and execute Python scripts at runtime. These scripts are constructed by interpolating data provided by the user or fetched from the web into code templates. This pattern of dynamic code generation is susceptible to injection attacks if the input data is not strictly validated.
[COMMAND_EXECUTION]: The skill's data collection script, scripts/fetch.py, explicitly disables SSL certificate verification using ssl._create_unverified_context(). This is a poor security practice that makes the skill's network communications vulnerable to man-in-the-middle (MITM) attacks, potentially allowing an attacker to intercept or modify the data retrieved from the web.
[SAFE]: The skill's network operations are directed at trusted and well-known services (Reddit, Algolia, DuckDuckGo, GitHub). It does not require sensitive user credentials, and its use of an optional GITHUB_TOKEN follows standard practices for public API access.
[SAFE]: The skill contains automated 'Self-QA' steps in SKILL.md to ensure the final output meets formatting requirements and avoids generic marketing language, demonstrating a design focused on data quality.

where-your-customer-lives