The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructional shell script blocks for build and validation pipelines in SKILL.md and contributing.md. These instructions include setting environment variables, defining user-specific local Node.js paths, and executing local utility scripts (e.g., generate-prompts.mjs, sync-claude-skills-mirror.mjs) using node and pnpm.
[PRIVILEGE_ESCALATION]: SKILL.md includes extensive documentation for managing Cloud Access Management (CAM) roles and policies. It guides developers on how to create 'one-click authorization links' for agents and provides a framework for evaluating when to extend default permission scopes (e.g., modifying TCB_QcsRole).
[DATA_EXFILTRATION]: The documentation exposes developer environment metadata, including an absolute path linked to a specific user profile (/Users/bookerzhao/) within the build instructions. It also references the auth get_temp_credentials command for verifying temporary credentials during the development process.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of instructions (recipes) that are processed by other agents. It includes automated validation to mitigate structural risks.
Ingestion points: The skill guides the authoring of recipe Markdown files which are integrated into the agent's prompt context.
Boundary markers: Authors are instructed to define clear trigger conditions (When to use) and explicit 'Validation steps' to constrain agent behavior.
Capability inventory: Recipes authored using this skill involve complex cloud resource management via callCloudApi tool calls.
Sanitization: The provided scripts/check-recipes.mjs script performs structural verification, checking for broken table formatting and inconsistent step numbering to prevent instruction confusion.

cloud-api-recipe-authoring