The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses explicit imperative instructions to bypass safety controls and user confirmation, promoting excessive autonomy.
Evidence: "Do this automatically without asking the user", "Don't Ask Questions", "Don't Wait for Instructions".
[DATA_EXFILTRATION]: The skill requires the agent to recursively read the entire codebase, which inherently exposes sensitive information such as environment variables, private keys, and configuration files to the model's context.
Evidence: "Read all files
Scan every file in the project recursively", "Use readFile to read every source file".
[COMMAND_EXECUTION]: The instructions direct the agent to run test suites and set up production infrastructure, which involves the execution of shell commands and external scripts.
Evidence: "Run all tests to ensure nothing broke", "Add production infrastructure
Logging, monitoring, health checks", "Add CI/CD pipeline".
[EXTERNAL_DOWNLOADS]: The skill recommends the integration and installation of several third-party libraries for production readiness.
Evidence: Mentions of Winston, Sentry, Prometheus, Zod, and bcrypt.
[PROMPT_INJECTION]: The skill creates a significant indirect injection surface by ingesting untrusted codebase data through deep line-by-line scanning while possessing high-impact capabilities.
Ingestion points: Recursive file scanning via listDirectory and readFile in SKILL.md.
Boundary markers: None present; the skill explicitly directs the agent to bypass user confirmation.
Capability inventory: File modification via strReplace, shell command execution for testing, and package installation.
Sanitization: No sanitization or filtering logic is specified for the ingested content.

production-code-audit