The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): Documentation in references/docs/guides/manual-installation.md instructs users to download binaries directly from GitHub releases and execute them (curl ... -o biome && chmod +x biome). While the source is a reputable project, executing unverified binaries bypasses standard package integrity checks.
PRIVILEGE ESCALATION (MEDIUM): The script scripts/scrape_biome_docs.py suggests using the --break-system-packages flag with pip, which allows overriding OS-level protections against global package installation. Additionally, installation guides recommend using chmod +x on downloaded binaries, granting them execution rights.
COMMAND_EXECUTION (LOW): The skill relies on executing various CLI commands through package managers (npx, pnpm, yarn, bun) and system utilities (jq, curl). This is consistent with the skill's purpose but requires shell access.
PROMPT_INJECTION (LOW): An indirect prompt injection surface exists as the skill provides a script to scrape external documentation from biomejs.dev for storage in the agent's reference library.
Ingestion points: scripts/scrape_biome_docs.py fetches data from biomejs.dev.
Boundary markers: Absent; scraped content is saved as raw markdown files without explicit delimiters or warnings to the agent.
Capability inventory: Extensive shell command execution capabilities provided by the skill's primary workflows.
Sanitization: The scraping script cleans formatting but does not filter for potential malicious instructions embedded in the documentation.

biome