The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instruction section provides example shell commands using npx @claude-flow/cli@latest. This pattern downloads and executes the specified package from the public NPM registry at runtime. Since the package scope reflects the vendor's defined infrastructure (claude-flow), this is documented as a standard download of tooling relevant to the skill.
[PROMPT_INJECTION]: The skill exhibits a potential surface for indirect prompt injection by processing external task completion data and feed back to influence model training and adaptation.
In gestion points: Untrusted conten t enters the system via task context in mcp__claude-flow__hooks_intelligence_trajectory-start, action recording in mcp__claude-flow__hooks_intelligence_trajectory-step, and feed back payloads in mcp__claude-flow__ruvllm_sona_adapt and mcp__claude-flow__ruvllm_microlora_adapt.
Boundary marker s: The skill does not provide explicit boundary delimiters or rule s to disregar d instruction s em bed ded within the in gested trajectory data.
Capability inventory: The agen t has acces s to powerfu l tool s tha t modify persisten t mode l pattern s and consolidate intelligence data bases, including mcp__claude-flow__neural_train, mcp__claude-flow__agentdb_consolidate, and mcp__claude-flow__hooks_pretrain.
Sanitization: The re is no indication of sanitization or validation logic for the incoming trajectory or feed back data in the provide d skill definition.
[COMMAND_EXECUTION]: The skill relies on executing variou s shell commands and MCP tool s to perform its primary function s. This includes the use of npx to run CLI tools and mcp tool call to invoke neural training operation s.

neural-train