The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to execute external code using the npx claude-flow@alpha command across multiple phase instructions.
Evidence: Found in SKILL.md (Phases 1 and 3) and PROCESS.md (Phases 1, 3, and 5).
Risk: Using npx to run unverified packages from a public registry at runtime can lead to remote code execution if the package or its dependencies are compromised.
[EXTERNAL_DOWNLOADS]: Fetches and executes the claude-flow package from the public NPM registry at runtime.
Evidence: Multiple shell command examples using npx to invoke the tool without pre-installation.
[COMMAND_EXECUTION]: Executes shell commands with potential for argument injection when processing user input.
Evidence: npx claude-flow@alpha hooks pre-task --description "Planning token budget for [task]" in PROCESS.md.
Risk: The skill interpolates the user-provided [task] description directly into a shell command argument. Without explicit instructions for sanitization or escaping shell metacharacters (e.g., ;, &, |), a malicious task description could be crafted to execute arbitrary shell commands.
[PROMPT_INJECTION]: Vulnerable to indirect prompt injection through untrusted task descriptions.
Ingestion points: The taskDescription variable in SKILL.md used for complexity estimation and phase decomposition.
Boundary markers: Absent. There are no delimiters (e.g., XML tags, triple backticks) or specific instructions to the agent to treat the user-provided task description as data rather than instructions.
Capability inventory: The agent has access to shell execution (npx), memory storage, and multi-agent coordination capabilities.
Sanitization: Absent. The skill logic uses regex for classification but does not validate or sanitize the input content before processing or interpolation into commands.

when-managing-token-budget-use-token-budget-advisor