The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements a "FORCE MODE" that automatically executes shell commands including git add -A, git commit, and git push without user confirmation. This bypasses typical safety checkpoints for code changes.
[PROMPT_INJECTION]: The instructions explicitly direct the agent to override standard confirmation protocols ("push ngay không hỏi", "FORCE MODE: push ngay không hỏi") based on a specific input prefix, reducing user oversight of autonomous actions.
[CREDENTIALS_UNSAFE]: The report generation logic specifically instructs the agent to extract and display information about required "secrets" and "environment variables" ("biến env cần set, secret cần có") found within configuration files.
[DATA_EXPOSURE]: The skill performs broad scans of the filesystem for sensitive configuration files (e.g., Jenkinsfile, fly.toml, cloudbuild.yaml) and reads their contents to extract deployment logic and environment targets.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests and processes untrusted data from multiple repository files (Jenkins, GitHub Actions, Docker, etc.) without explicit boundary markers or sanitization, potentially allowing files within the repository to influence agent behavior.

ci-cd