The Agent Skills Directory

[METADATA_POISONING]: The skill's YAML frontmatter identifies the author as "github" and provides links to official GitHub repositories, but the skill is actually authored by "pmndrs". This misrepresentation of the source is deceptive.
[EXTERNAL_DOWNLOADS]: The skill directs the agent to install the gh-stack extension from the official github/gh-stack repository, which is a reference to a well-known service's official resource.
[COMMAND_EXECUTION]: The skill instructs the agent to execute various GitHub CLI (gh) and Git commands to manage local and remote code branches.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes pull request metadata (such as titles and bodies) and branch names from external sources, creating a potential surface for indirect instruction injection.
Ingestion points: Pull request metadata and branch names accessed via gh stack view --json and gh stack checkout.
Boundary markers: No explicit markers are used to delimit or neutralize instructions embedded in PR metadata.
Capability inventory: The skill possesses capabilities for file system modification (via Git), branch management, and network communication (via GitHub CLI).
Sanitization: The instructions do not include steps to sanitize or validate external PR content before it is processed by the agent.

gh-stack