The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's strategy documentation in references/strategies.md provides shell command templates that incorporate variable parts of a URL, such as <handle>, <rkey>, <instance>, or <id>, into command strings. If these parameters are not rigorously escaped before the agent executes the shell commands, it creates an opportunity for command injection.
[DATA_EXFILTRATION]: To fulfill its primary function, the skill transmits user-provided social media URLs to external services, including ScrapeCreators and the well-known platform Apify. This involves sharing potentially private URL data with third-party infrastructure.
[EXTERNAL_DOWNLOADS]: The skill fetches data from various social media APIs and optionally downloads media files to the user's local directory at ~/Documents/social-fetches/. These operations are conducted using standard tools like curl and agent-browser.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) due to its data processing pipeline:
Ingestion points: Fetches untrusted text and metadata from platforms such as X, Reddit, and LinkedIn as specified in SKILL.md.
Boundary markers: The skill lacks clear delimiters or instructions to prevent the agent from being influenced by malicious content embedded within the fetched social media posts.
Capability inventory: The execution environment allows for shell command execution and local file system modifications.
Sanitization: There is no documented process for sanitizing or escaping the retrieved content before it is returned and used by other skills.

social-fetch