The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Retrieves maintenance scripts such as updateBaseImages.sh and createPR.sh from the vendor's internal GitLab instance (gitlab.cee.redhat.com).
[EXTERNAL_DOWNLOADS]: Fetches product lifecycle information and RPM repository data from official Red Hat and OpenShift domains (access.redhat.com, mirror.openshift.com).
[EXTERNAL_DOWNLOADS]: Downloads Node.js headers from the official distribution site at nodejs.org to align development checkouts with builder images.
[REMOTE_CODE_EXECUTION]: Installs and executes the rpm-lockfile-prototype utility from the konflux-ci GitHub organization to manage RPM dependencies.
[REMOTE_CODE_EXECUTION]: Downloads and executes utility scripts from the vendor's private GitLab repository to perform automated image tagging and PR creation.
[COMMAND_EXECUTION]: Utilizes system tools including git, gh, glab, skopeo, jq, and curl to manage local clones, interact with remote registries, and coordinate pull request workflows.
[DYNAMIC_EXECUTION]: Invokes container runtimes (podman or docker) to run node --version and go version from UBI builder images to extract version strings for toolchain synchronization.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external container registries and mirror sites which could contain malicious instructions. The risk is mitigated by explicit version regex validation and a mandatory manual approval gate for all write operations.
Ingestion points: Container registry metadata (image tags and comments), OpenShift mirror listings, and Red Hat Product Life Cycle API responses.
Boundary markers: The skill requires following a 'mutation-gate' protocol where every command and target must be approved by the user before execution.
Capability inventory: Possesses capabilities to read/write files, execute shell commands, manage git branches, and create PRs/MRs.
Sanitization: Implements strict regular expression checks on version tags and uses structured JSON parsing for external API payloads.

rhdh-base-images