The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes legacy source code as input during its initial requirements gathering phase (Phase 0). There is an inherent risk that untrusted data or instructions embedded in the legacy code could influence the agent's behavior.
Ingestion points: Legacy source code from grafana-cloud-cli is read and summarized in Phase 0.
Boundary markers: The workflow implements a strict five-phase pipeline where the Lead agent must obtain explicit user approval at Phase 1 (Design), Phase 2 (Spec), and Phase 4 (Verification) before proceeding.
Capability inventory: The skill has the capability to write Go source files, execute local binaries (bin/gcx), and run build/test commands via mise.
Sanitization: While no automated sanitization of source code is described, the mandatory human review checkpoints for every design document and the final comparison report serve as the primary security control.
[DYNAMIC_EXECUTION]: The skill's primary function is to generate Go source code for resource adapters, clients, and CLI commands. This code is subsequently compiled and executed using mise run all during the Build and Verification phases. While this involves dynamic generation and execution of code, it is the core intended purpose of the developer-focused skill and is subject to comprehensive unit testing, linting, and human verification.
[COMMAND_EXECUTION]: The instructions require executing the local bin/gcx binary and other development tools to interact with live Grafana instances to perform smoke tests and verify configuration (Phase 4). These operations are restricted to the local environment and targeted Grafana contexts provided by the user, representing standard development workflow operations.

migrate-provider