The Agent Skills Directory

[PROMPT_INJECTION]: No patterns of instruction overrides, safety filter bypasses, or role-play injection (such as DAN) were detected. The skill instructions are focused on adherence to development workflows.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill utilizes git push and the GitHub CLI (gh) to manage contributions. These operations are targeted at GitHub, which is a recognized and trusted service. No sensitive files or hardcoded credentials were found, and network operations are consistent with the skill's purpose.
[REMOTE_CODE_EXECUTION]: Local project scripts are executed using bun run. These are internal build and verification tools. No remote code execution patterns, such as piping external URLs to a shell, were identified.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The agent is instructed to read GitHub Issues and PR comments via the gh tool to understand tasks and rationale.
Boundary markers: None are explicitly defined in the instructions to separate untrusted issue content from instructions.
Capability inventory: The skill has capabilities for local file system operations (git), project script execution (bun), and network interaction (gh).
Sanitization: The skill relies on standard LLM safety guardrails; there is no explicit sanitization of external issue data provided in the scripts.
[DYNAMIC_CONTEXT_INJECTION]: No instances of the !command syntax or dynamic context injection were found in the provided files.
[SAFE]: All external resource references and command executions are standard for software development workflows and align with the stated purpose of the skill.

voyager-contribute