The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads a pre-compiled binary from an external GitHub repository (github.com/rtk-ai/rtk) and extracts it directly into /usr/local/bin using curl and tar.
[PRIVILEGE_ESCALATION]: The skill modifies the agent's global configuration file (~/.claude/settings.json) by executing rtk init -g. This expands the tool's influence over the agent's core operational environment.
[PERSISTENCE]: By hooking into the agent's global settings, the skill ensures that its modifications persist across different sessions and may influence future agent interactions.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from potentially untrusted project files to seed the wiki metadata.
Ingestion points: Reads content from README.md, package.json, and go.mod (referenced in W05 and Step 5).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used during interpolation.
Capability inventory: Includes shell command execution (curl, tar, rtk), filesystem writes, and modification of agent settings.
Sanitization: No sanitization or validation of the ingested content is described before it is used to generate wiki pages or project requirements.
[COMMAND_EXECUTION]: Executes system-level commands including kernel checks (uname -r), directory manipulation (mkdir -p), and binary installation to restricted paths.

new-project-setup