The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download and execute code from a remote repository at github.com/SeemSeam/claude_codex_bridge.git during the installation process.
[COMMAND_EXECUTION]: The installation instructions for Windows utilize powershell -ExecutionPolicy Bypass, which explicitly overrides system security settings to run an unverified script.
[REMOTE_CODE_EXECUTION]: The ccb update command downloads and executes remote updates, which allows for the runtime modification of the tool's behavior from an external source.
[COMMAND_EXECUTION]: The provided Python integration examples use subprocess.run to execute CLI commands based on agent input. If the agent_name or query parameters are sourced from untrusted data, this could lead to command injection.
[DATA_EXFILTRATION]: The configuration system encourages storing API keys and base URLs in a local file (.ccb/ccb.config). While it advises using environment variables, the framework's support for direct key placement increases the risk of credential exposure if files are accidentally shared.
[PROMPT_INJECTION]: The skill establishes a shared memory surface in .ccb/ccb_memory.md used by multiple agents.
Ingestion points: All agents read and write to .ccb/ccb_memory.md to maintain project context.
Boundary markers: No specific delimiters or instruction-ignore guards are defined for the shared memory content.
Capability inventory: The skill allows subprocess execution, file system writes, and shell access via tmux panes.
Sanitization: There is no evidence of sanitization or validation of data before it is written to or read from the shared memory file.

seemseam-ccb-multi-agent-cli