The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation and scripts extensively use 'npx claude-flow' to initialize swarms and spawn agents. This command downloads and executes code from the npm registry at runtime. As 'claude-flow' is not a verified package from a trusted vendor, this represents a remote code execution risk of unverified software.
[COMMAND_EXECUTION]: The skill provides numerous high-privilege shell commands for the agent to execute, including 'swarm init', 'agent spawn', and 'workflow execute'. These commands allow the creation and control of multiple sub-processes and specialized agents within the environment.
[DATA_EXFILTRATION]: The orchestration workflow sends sensitive project data—including market analysis, business models, and technical specifications—to an external 'memory system' using 'npx claude-flow memory store'. The storage backend and security protocols for this data are managed by the third-party tool and are not explicitly defined, creating a risk of unauthorized data exposure.
[EXTERNAL_DOWNLOADS]: The skill's functionality is dependent on fetching the 'claude-flow' package from a remote repository (npm) at runtime without integrity verification or version pinning.
[PROMPT_INJECTION]: The workflow is vulnerable to indirect prompt injection due to its multi-agent data flow. 1. Ingestion points: Data is ingested into agent contexts using 'npx claude-flow memory retrieve' (SKILL.md). 2. Boundary markers: No delimiters or instructions to ignore embedded commands are present in the provided templates. 3. Capability inventory: The agents have extensive capabilities including shell execution via npx and complex orchestration (SKILL.md). 4. Sanitization: There is no evidence of sanitization or validation of the data retrieved from memory before it is processed by subsequent agents.

when-releasing-new-product-orchestrate-product-launch