The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/check-rtk.sh provides a command to install the RTK utility by piping a remote script directly into a shell (curl -fsSL ... | sh). This is a high-risk execution pattern that runs unverified code from an external server.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download and execution of software from the rtk-ai GitHub organization, which is an untrusted third-party source not included in established safe-lists.
[COMMAND_EXECUTION]: Instructions in SKILL.md and references/COMMAND_MAP.md direct the agent to proxy sensitive terminal operations (such as aws sts, kubectl, and psql) through the rtk binary. This introduces a supply-chain risk where a compromised binary could intercept credentials or modify command behavior.
[DATA_EXFILTRATION]: By promoting the use of a third-party command wrapper for cloud provider and database interactions, the skill creates a potential point of interception for sensitive credentials and environment variables.
[PROMPT_INJECTION]: The skill processes terminal output through the rtk tool to summarize it. If an attacker can control the output of a command being wrapped (e.g., via a malicious filename or log entry), they could attempt an indirect prompt injection to influence the agent's behavior, although this is mitigated by the tool's intended purpose of reducing output.

rtk-token-saver