The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the sidecar CLI tool to perform various operations, including listing shells and agents, creating new shells, and prompting agents. These commands are executed within the agent's environment. Evidence includes commands such as sidecar shell list, sidecar agent prompt, and sidecar create shell.\n- [INDIRECT_PROMPT_INJECTION]: The skill facilitates a workflow where the agent ingests untrusted data from a secondary agent's output and uses it to drive further actions, creating a vulnerability surface.\n
Ingestion points: The sidecar agent read command (in SKILL.md) is used to read screen content, scrollback, and transcripts from target shells.\n
Capability inventory: The orchestrating agent can execute sidecar agent prompt, sidecar agent send-keys, and sidecar agent broadcast based on its interpretation of the read data.\n
Boundary markers: No explicit instruction delimiters or boundary markers (e.g., XML tags or special tokens) are mentioned in the skill to isolate ingested data from agent instructions.\n
Sanitization: The skill relies on procedural human-in-the-loop style review by the agent ("Read the screen, decide, then answer") rather than automated sanitization or filtering of the external content.

coordinate-agents