The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions direct the agent to read various repository files to determine which manual steps and credentials are required for the wizard.
Ingestion points: The agent reads project files like README, .env.example, docker-compose.yml, and .github/workflows/* as specified in SKILL.md under the 'Scope the procedure' section.
Boundary markers: There are no explicit boundary markers or instructions to ignore potential commands embedded within these project files.
Capability inventory: The generated script includes functions to write to local files (write_env), interact with the GitHub CLI (set_secret, set_var), and launch the user's browser (open_url).
Sanitization: The skill includes a human-in-the-loop requirement where the agent must show the ordered list of stages to the user for confirmation before generating the script.
[DYNAMIC_EXECUTION]: The skill is designed to generate executable bash scripts based on a provided template and AI-authored logic.
Evidence: SKILL.md instructions (Step 3: Author the wizard) guide the agent to copy template.sh and populate it with custom stages to be executed by the user.
[COMMAND_EXECUTION]: The skill encourages the use of local command-line tools for verification and execution of the generated scripts.
Evidence: SKILL.md (Step 4: Verify and hand off) suggests running bash -n, shellcheck, and chmod +x on the generated files.

wizard