The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The methodology processes untrusted inputs through various agent modes, creating a potential surface for indirect prompt injection.\n
Ingestion points: User-provided task_description strings and external content retrieved by the researcher mode in SKILL.md.\n
Boundary markers: No explicit instruction delimiters or containment markers are defined for the input data in the described workflows.\n
Capability inventory: The framework includes the coder mode for file modification, researcher for network access, and orchestrator for agent delegation as described in SKILL.md.\n
Sanitization: The methodology does not specify validation or sanitization routines for external data or task descriptions before they influence agent behavior.\n- [EXTERNAL_DOWNLOADS]: The skill utilizes the npx utility to download and execute the claude-flow package from the NPM registry.\n- [REMOTE_CODE_EXECUTION]: The npx command is used to fetch and run the claude-flow package from the official NPM registry.\n- [COMMAND_EXECUTION]: The skill methodology involves executing CLI commands to run modes, manage swarms, and trigger hooks.\n- [DYNAMIC_EXECUTION]: The framework dynamically spawns specialized agents and orchestrates multi-agent swarms based on task requirements and optimized topologies.

qe-sparc-methodology