The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill installs software from external sources, specifically fetching the rdt-cli package directly from a GitHub repository (public-clis/rdt-cli.git) via pipx instead of a standard registry.
[REMOTE_CODE_EXECUTION]: The skill implements a remote update and installation process. Commands like agent-reach install --env=auto and agent-reach check-update fetch and potentially execute scripts or instructions from remote GitHub locations (raw.githubusercontent.com).
[COMMAND_EXECUTION]: The skill frequently invokes shell commands and subprocesses to manage third-party tools such as yt-dlp, gh, twitter-cli, rdt-cli, and bilibili-cli, including scripts stored in the user's home directory (~/.agent-reach/).
[CREDENTIALS_UNSAFE]: The skill instructs users to manually provide and set sensitive authentication data, such as TWITTER_AUTH_TOKEN and TWITTER_CT0, to enable access to protected platform features.
[COMMAND_EXECUTION]: Dynamic code execution is utilized through python3 -c for tasks like RSS feed parsing and within provided Python code snippets for platform interactions.
[DATA_EXFILTRATION]: The skill is susceptible to indirect prompt injection due to the high volume of untrusted content it processes from the internet.
Ingestion points: Data is fetched via Exa Search (references/search.md), Jina Reader (references/web.md), social media feeds from Twitter, Reddit, and XHS (references/social.md), and video subtitles (references/video.md).
Boundary markers: There are no explicit markers or instructions used to isolate untrusted fetched content from the agent's core logic.
Capability inventory: The skill possesses significant capabilities including shell command execution, network access, and file system writes (detailed in SKILL.md and references/dev.md).
Sanitization: The skill lacks mechanisms to sanitize or validate external content before it is processed by the AI agent.

agent-reach