The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands, including opencode CLI calls for model execution and curl for interacting with a local API server. It also uses a one-line Python script to parse JSON responses.
[DATA_EXFILTRATION]: The skill includes instructions to perform network operations against http://localhost:4096. These requests are used for connectivity heartbeats and API interaction with a local service. Localhost is a whitelisted domain, and no external data exfiltration was detected.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by instructing the agent to interpolate {constructed_prompt} (which may contain untrusted data from external sources) into shell command arguments or API request bodies.
Ingestion points: The {constructed_prompt} variable is interpolated into opencode run commands and HTTP POST payloads in SKILL.md.
Boundary markers: No explicit boundary markers or 'ignore' instructions for the interpolated content are defined in these reference files.
Capability inventory: The skill utilizes subprocess execution and local network access.
Sanitization: There is no evidence of input validation or sanitization for the interpolated data within the skill's instructions.

bridge-opencode