The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes various repository files to define the scope of the manual procedure, creating a potential surface for indirect prompt injection from untrusted repository content.
Ingestion points: The skill instructions (SKILL.md) direct the agent to read .env, .env.example, README, Docker Compose files, framework configurations, and GitHub workflow files (.github/workflows/*) to identify necessary variables and secrets.
Boundary markers: No specific delimiters or safety instructions are used to separate content read from these files from the agent's instructions.
Capability inventory: The generated wizard scripts have the capability to write to local files (write_env), open URLs in the user's browser (open_url), and interact with GitHub repository secrets and variables (set_secret, set_var) via the gh CLI.
Sanitization: The skill does not explicitly define sanitization or validation routines for data extracted from repository files before it is used to author the wizard stages.
[DYNAMIC_EXECUTION]: The skill generates and facilitates the execution of bash scripts authored at runtime based on a provided template.
Evidence: The skill uses scripts/template.sh as a base to author ephemeral scripts (wizard-<slug>.sh) and provides the user with commands to make these scripts executable and run them.
[COMMAND_EXECUTION]: The generated wizard scripts execute local system commands and interact with external services through official tools.
Evidence: The open_url function in scripts/template.sh utilizes system utilities like wslview, explorer.exe, xdg-open, and open to trigger browser activity. Additionally, the set_secret and set_var functions execute the gh (GitHub CLI) tool to manage repository configurations, which is a standard and expected behavior for a provisioning utility.

wizard