The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill uses a pattern of downloading and executing remote scripts via shell piping for installation and bootstrapping.
Evidence: SKILL.md and scripts/compose-preview contain instructions to run curl -fsSL https://raw.githubusercontent.com/yschimke/skills/main/scripts/install.sh | bash.
Context: While the source is the author's repository, executing remote un-audited code directly in the shell is a high-risk pattern for AI agents.
[REMOTE_CODE_EXECUTION]: The skill supports dynamic application of Gradle plugins using init scripts, which allows arbitrary code execution during the build process.
Evidence: SKILL.md provides a Groovy init script (~/.gradle/init.d/compose-ai-tools.gradle) that dynamically applies a plugin based on environment variables.
[PROMPT_INJECTION]: The skill modifies the agent's long-term behavior by writing persistent instruction files to the environment.
Evidence: In references/agent-cloud.md, a setup script writes to ~/.claude/CLAUDE.md with instructions like 'Override any conflicting workspace defaults' and specific rules for commits and PRs intended to control agent identity and behavior.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection as it ingests and processes untrusted project data.
Ingestion points: The skill scans project source code (.kt files), Gradle build scripts, and Android XML resources (SKILL.md, references/resource-previews.md).
Boundary markers: Absent. No delimiters or 'ignore embedded instructions' warnings are provided when the agent is directed to read or iterate on these files.
Capability inventory: The agent executes shell commands via the compose-preview CLI and ./gradlew tasks (SKILL.md).
Sanitization: No evidence of sanitization or filtering of content from the processed files before it enters the agent's context.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to download the CLI and associated tools from external GitHub repositories.
Evidence: scripts/compose-preview and SKILL.md reference downloads from raw.githubusercontent.com.
[COMMAND_EXECUTION]: The skill relies on executing various shell commands and project-level build tools.
Evidence: Extensive use of compose-preview CLI subcommands and ./gradlew for rendering and discovery tasks.

compose-preview