The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various development commands to perform and verify the upgrade, including yarn backstage-cli versions:bump, yarn backstage-cli versions:migrate, yarn tsc, yarn build, and yarn test. These are standard operations for the stated purpose of upgrading a software project.
[EXTERNAL_DOWNLOADS]: Fetches release manifests from https://versions.backstage.io and retrieves documentation/changelogs from the official Backstage GitHub repository. These sources are official and well-known within the Backstage ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data and has exploitable capabilities.
Ingestion points: Reads external changelogs and release notes from github.com/backstage/backstage (as seen in references/fix-breaking-changes.md).
Boundary markers: No explicit delimiters or instructions are provided to the agent to disregard instructions potentially embedded within the external changelog text.
Capability inventory: The agent can execute shell commands (yarn, grep, cat) and modify project files (package.json, source code).
Sanitization: No sanitization or validation of the fetched changelog content is described before the agent processes it for decision-making.
[DATA_EXFILTRATION]: The skill reads project configuration files like package.json and backstage.json to identify current versions. While this is necessary for the upgrade, this data access combined with the network access used by the CLI tools represents a theoretical exfiltration path, though no malicious behavior was detected.

backstage-upgrade