The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): File references/harness-mem.md contains a piped shell execution pattern (curl -fsSL https://bun.sh/install | bash) to install the Bun runtime. This executes remote code directly into the user's shell environment without integrity verification.
[REMOTE_CODE_EXECUTION] (HIGH): File references/remotion-setup.md utilizes npx create-video@latest to download and execute code from a remote npm package at runtime.
[EXTERNAL_DOWNLOADS] (MEDIUM): Multiple files (references/dev-tools-setup.md, references/lsp-setup.md) suggest installing numerous global and local packages via npm, pip, and brew without version pinning or checksum validation, posing a supply chain risk.
[COMMAND_EXECUTION] (HIGH): The skill frequently executes local bash scripts (e.g., scripts/analyze-project.sh, scripts/codex-setup-local.sh, scripts/localize-rules.sh) that are not provided within the skill bundle, meaning their behavior is unverifiable and potentially malicious.
[COMMAND_EXECUTION] (MEDIUM): references/2agent-setup.md performs chmod +x on a dynamically created/copied script (.claude/scripts/auto-cleanup-hook.sh) to allow its execution as a hook.
[DATA_EXPOSURE] (LOW): While the skill does not hardcode secrets, it instructs the user to set up environment variables for sensitive API keys (ANTHROPIC_API_KEY, AIVIS_API_KEY, GOOGLE_AI_API_KEY) and configures CI workflows that utilize these secrets.
[PROMPT_INJECTION] (LOW): references/2agent-setup.md explicitly recommends that users do not disable bypassPermissionsMode, which allows the agent to execute tools without manual confirmation, significantly increasing the impact of a potential prompt injection attack.

setup