The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to implement a self-update routine by fetching a new version of its instruction file from https://deepapi.co/deepapi-skill/SKILL.md and overwriting the local file when versions differ.
[REMOTE_CODE_EXECUTION]: By directing the agent to overwrite its local instruction set with content fetched from a remote URL, the skill facilitates a vector for remote instruction injection.
[COMMAND_EXECUTION]: To retrieve credentials, the instructions require the agent to execute shell commands that source local configuration files, specifically source ~/.deepapi/env for bash/zsh and . "$HOME/.deepapi/env.ps1" for PowerShell.
[CREDENTIALS_UNSAFE]: The skill explicitly targets sensitive credential files located in the user's home directory (~/.deepapi/env) to obtain API keys and environment variables.
[PROMPT_INJECTION]: The skill is exposed to indirect prompt injection risks due to its core function of scraping untrusted data from multiple social media platforms and web searches.
Ingestion points: Multiple endpoints for scraping LinkedIn, GitHub, X/Twitter, and general websites.
Boundary markers: The instructions do not define delimiters or specific guidelines to prevent the agent from following instructions embedded in the scraped data.
Capability inventory: The agent possesses capabilities to send emails, post to X/Twitter, and deploy web pages using content retrieved from external sources.
Sanitization: No explicit validation, filtering, or sanitization steps are provided for the agent to perform on the ingested content before downstream processing.

deepapi