The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from the user's local source code repository, creating a surface for indirect prompt injection.
Ingestion points: The agent is instructed to read local source code, workload.yaml descriptors, and CI/CD configuration files (like .github/workflows) to determine the application shape and deployment requirements, as seen in getting-started.md and deploy-prebuilt-image.md.
Boundary markers: Instructions include explicit guardrails requiring user confirmation before the agent performs any git actions (commit, push, PR) or destructive platform operations (e.g., delete_component).
Capability inventory: The skill possesses extensive capabilities through the openchoreo-cp MCP server to manage platform resources (CRUD operations on components, workloads, and secrets) and uses local git and gh tools for repository management.
Sanitization: No specific sanitization or prompt-wrapping techniques are provided to mitigate instructions that might be embedded in the analyzed code.
[COMMAND_EXECUTION]: The skill uses standard development tools to perform repository-related tasks.
Evidence: Files like references/recipes/build-from-source.md and references/recipes/deploy-prebuilt-image.md guide the agent to use git and gh (GitHub CLI) for staging, committing, pushing code, and managing Pull Requests. These actions are governed by instructions requiring user approval.
[NO_CODE]: No executable script files (e.g., .py, .js, .sh) are distributed with this skill; it consists entirely of instructional markdown and configuration templates.

openchoreo-developer