The Agent Skills Directory

[PROMPT_INJECTION]: The instructions employ manipulative reinforcement techniques, such as claiming a formatting error is a "catastrophic error that may result in human death," to pressure the AI into strict compliance with output rules.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). It retrieves commit messages from the project history and analyzes them to generate a changelog. Malicious commit messages authored by external contributors could contain hidden instructions intended to hijack the agent's behavior during the analysis phase.
[COMMAND_EXECUTION]: The skill executes multiple shell commands, including git operations and generic project build commands. Building a project is a significant attack surface, as it can trigger arbitrary code execution via build hooks (e.g., npm scripts, Makefiles, or setup.py) if the repository content is untrusted.

bump-version