The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documentation and scripts utilize npx to download and execute the claude-flow@alpha package. This is the primary method for using the vendor's orchestration tool and involves remote code execution at runtime.
[COMMAND_EXECUTION]: The skill performs shell command execution to manage the lifecycle of agents and the execution of the data pipeline. Specifically, it uses subcommands like agent spawn to create sub-processes and pipeline execute to run workflows.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an Indirect Prompt Injection surface due to its design of chaining multiple agent stages together.
Ingestion points: Data enters the agent context through input files like initial-data.json and flows between stages via memory and streaming channels as described in SKILL.md.
Boundary markers: There are no explicit boundary markers or instructions to ignore embedded commands within the passed data mentioned in the workflow or scripts.
Capability inventory: The pipeline orchestration involves capabilities such as spawning new agent processes (agent spawn) and executing multi-stage logic (pipeline execute) across SKILL.md and PROCESS.md.
Sanitization: No evidence of data sanitization, escaping, or schema validation was found to prevent adversarial content in one agent's output from influencing the behavior of downstream agents.

when-chaining-agent-pipelines-use-stream-chain