The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides the hitch send-keys command, which allows the agent to send arbitrary input and shell commands to any open terminal session. This is a high-privilege capability that can be used to execute unintended code if the agent is manipulated.
Evidence: hitch send-keys -t 1 C-u "npm test" Enter in SKILL.md.
[DATA_EXFILTRATION]: The skill allows the agent to read terminal history and current output using hitch context and hitch capture. This could lead to the exposure of sensitive information (e.g., credentials, secrets, or private data) that might be visible in the user's terminal or log files.
Evidence: hitch context, hitch context --all, and hitch capture -t <terminal> in SKILL.md.
[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection because it ingests untrusted data from terminal outputs which could contain malicious instructions.
Ingestion points: hitch context and hitch capture in SKILL.md read terminal output into the agent's context.
Boundary markers: Absent. There are no instructions or markers to help the agent distinguish between legitimate terminal output and potential malicious instructions embedded in that output.
Capability inventory: The agent has access to hitch send-keys (command execution) and potentially other native tools like file writing or network requests.
Sanitization: Absent. There is no mention of filtering or sanitizing terminal output before it is processed by the agent.

hitch