The Agent Skills Directory

[COMMAND_EXECUTION]: The skill extensively uses subprocess and pty modules to interact with the system and drive the interactive installer.\n
scripts/wizard-run.py uses pty.fork() and os.execvpe() to run bash nanoclaw.sh in a controlled terminal environment.\n
It executes various system tools including docker, systemctl, launchctl, and the onecli utility to verify installation status.\n- [DYNAMIC_EXECUTION]: The skill generates and executes code at runtime to facilitate testing and deployment.\n
scripts/proxmox-wizard.py bundles Python scripts as Base64-encoded strings within a generated Bash script, which is then executed on remote guest machines via python3 -.\n
scripts/wizard-run.py dynamically creates a TypeScript (.mts) file and executes it using pnpm exec tsx to verify credentials through the product's internal adapters.\n- [EXTERNAL_DOWNLOADS]: The skill fetches dependencies and product payloads during the setup and testing phases.\n
scripts/wizard-install.sh installs Python dependencies from requirements.txt using pip with hash verification.\n
scripts/wizard-run.py may dynamically install the @openai/codex package via npm if a host-side version is missing or incompatible.\n- [REMOTE_CODE_EXECUTION]: The skill clones and executes code from remote repositories based on user-supplied parameters.\n
scripts/proxmox-wizard.py performs git fetch operations to retrieve provider payloads and makes the wizard install them through the product's normal registry remote.\n- [CREDENTIALS_UNSAFE]: The skill requires sensitive API keys to be provided via local files (--credential-file).\n
It reads these credentials into memory to simulate user input during the wizard process and to identify values that must be redacted from logs.\n- [DATA_EXFILTRATION]: The skill accesses sensitive information to perform its diagnostic and sanitization tasks.\n
scripts/wizard-run.py reads local .env files and process environment data from /proc/[pid]/environ to locate secrets for the Redactor utility.\n
While this is performed to sanitize logs before export, the capability to read arbitrary environment secrets is a significant data access vector.\n- [INDIRECT_PROMPT_INJECTION]: As an automated terminal driver, the skill is susceptible to indirect prompt injection.\n
scripts/wizard-run.py parses output from the installer process to identify prompts and submit answers. If the driven process (or a fetched payload) outputs maliciously crafted text that mimics expected prompts, it could influence the driver's actions.\n
Ingestion points: Terminal output captured from the nanoclaw.sh process and interpreted by the pyte emulator in scripts/wizard-run.py.\n
Boundary markers: The driver uses specific regex patterns and terminal state checks (e.g., DEVICE_CODE_PROMPT, CLAUDE_AUTH_URL) to identify legitimate interaction points.\n
Capability inventory: The skill has broad capabilities including subprocess execution, file system access, and Docker management across multiple scripts.\n
Sanitization: Implements a Redactor class that uses regex and whitespace-aware matching to scrub secrets from exported terminal transcripts and logs.\n- [PRIVILEGE_ESCALATION]: The skill performs actions that require elevated permissions to manage system dependencies.\n
scripts/wizard-install.sh uses sudo apt-get to install the python3-venv package if it is missing from the environment.

e2e-wizard