The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses several command-line tools to interact with the repository and GitHub API, including gh (GitHub CLI), git, pre-commit, bun, and a local repository script called ods.
Evidence includes commands like gh pr list, gh pr checkout, git rebase, and ods audit.
The commands are scoped to the onyx-dot-app/onyx repository, which matches the vendor context.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources (pull request titles, bodies, and release notes) to categorize PRs and generate summaries for the user.
Ingestion points: Data is fetched via gh pr list and gh pr view (implied by reading release notes in PR bodies).
Boundary markers: None explicitly defined for PR content interpolation.
Capability inventory: The agent can execute shell commands (Bash), perform file system operations via git/gh, and trigger repository-defined scripts via bun install.
Sanitization: No explicit sanitization of PR body text is mentioned, though the skill relies on the user for final confirmation.
The risk is mitigated by the skill's default filter for PRs authored by app/dependabot.
[REMOTE_CODE_EXECUTION]: The skill performs actions that can trigger the execution of code defined within the repository configuration.
Patterns: Running bun install on a PR branch can trigger postinstall scripts or other lifecycle hooks defined in package.json.
Mitigation: This behavior is standard for dependency management and is performed on branches ostensibly managed by Dependabot and the onyx-dot-app organization.

merge-dependabot-prs