The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements an activity monitoring system that processes untrusted data to generate suggestions, which is a classic surface for indirect prompt injection.
Ingestion points: Monitors file extensions, file paths, shell command history, error patterns from tool outputs, and project configuration files (e.g., package.json).
Boundary markers: No explicit delimiters or boundary markers are defined to isolate untrusted tool outputs from the suggestion logic.
Capability inventory: The skill facilitates the execution of various commands and tools through suggestion patterns, including automated build fixes and deployment scripts.
Sanitization: The instructions do not define any sanitization or validation of the ingested signals before they influence agent behavior.
[EXTERNAL_DOWNLOADS]: The skill suggests the installation of external software packages to provide specific functionality.
Evidence: Recommends npm install developer-icons for tech-related visual assets.
Reference: Points to a documentation site at https://xandemon.github.io/developer-icons/ for icon browsing.
[COMMAND_EXECUTION]: The skill generates and suggests executable shell commands based on the current project context and detected patterns.
Evidence: Provides templates for commands such as npm test, git push, docker build, and various project-specific 'slash commands' like /build-fix and /deploy.

contextual-assist