The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines a shell command template that interpolates user-provided arguments (MODE, DATE, LEVEL) directly into an npx execution string without sanitization or shell escaping. This pattern is vulnerable to command injection, allowing an attacker to execute arbitrary shell commands by providing malicious input in these fields.
[REMOTE_CODE_EXECUTION]: The skill fetches and executes the @claude-flow/plugin-agent-federation package from the NPM registry at runtime using npx. The use of the @latest version tag and a package source not recognized as a trusted vendor introduces a risk of supply chain attacks or execution of malicious updates.
[PROMPT_INJECTION]: The skill reads and summarizes federation audit logs, which are considered untrusted data sources. The absence of boundary markers or instructions to isolate log content creates a surface for indirect prompt injection, where malicious instructions embedded in the logs could manipulate the agent's summarization logic.
Ingestion points: Federation audit logs fetched via the ruflo-federation binary in Step 2.
Boundary markers: None present; the logs are processed directly into the agent's context for summarization.
Capability inventory: Bash (via npx), Read, Grep.
Sanitization: No filtering, validation, or escaping is performed on the log data before it is processed by the agent.

federation-audit