The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources including upstream provider code, tool logs, and replacement configurations, which presents a potential vulnerability surface.
Ingestion points: Processes output from upgrade-provider logs, the upstream git submodule, and provider/replacements.json.
Boundary markers: There are no specific delimiter-based boundary markers used when the agent reads these external data sources.
Capability inventory: The skill possesses the ability to perform git commits and pushes, gh PR modifications, and go code compilation/execution.
Sanitization: The skill relies on mandatory human review checkpoints (e.g., auditing replacements.json and reviewing diffs before pushing) rather than programmatic sanitization to mitigate risks from untrusted upstream content.
[COMMAND_EXECUTION]: The skill uses various command-line utilities for repository management and PR automation, such as git, gh, and upgrade-provider. These operations are well-scoped to the task of provider maintenance.
[DYNAMIC_EXECUTION]: The skill performs runtime environment setup and compilation using eval "$(mise env)", go test, and make tfgen. These are standard activities for the intended primary purpose of developing and building Pulumi providers.

pulumi-upgrade-provider