The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the installation and execution of code from external repositories via the npx skills add command. This pattern allows the agent to fetch and integrate executable skills from user-provided or context-derived URLs (e.g., https://github.com/username/repo-name).
[COMMAND_EXECUTION]: The skill contains platform-specific scripts (Bash and PowerShell) for relocating files between ~/.agents/skills/ and ~/.gemini/config/skills/.
Evidence (Linux/macOS): mkdir -p "$HOME/.gemini/config/skills/" && [ -d "$HOME/.agents/skills" ] && for d in "$HOME/.agents/skills"/*/; do [ -d "$d" ] && rm -rf "$HOME/.gemini/config/skills/$(basename "$d")" && mv "$d" "$HOME/.gemini/config/skills/"; done
Evidence (Windows): Uses Remove-Item -Recurse -Force and Move-Item -Force to manage directories.
[EXTERNAL_DOWNLOADS]: The skill triggers downloads from external sources, specifically GitHub repositories, to add or update agent capabilities.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it ingests and processes SKILL.md files from untrusted external repositories.
Ingestion points: The agent reads the first 10 lines of SKILL.md from downloaded repositories in Step 5 (Verification Loop).
Boundary markers: Absent; the skill does not use delimiters or instructions to ignore embedded content within the ingested files.
Capability inventory: The skill has access to shell execution (run_command), package installation (npx), and file write capabilities (skills.json).
Sanitization: Absent; there is no evidence of validation or filtering for the content read from external skill files.

deploy-skills-in-antigravity