The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes role-play instructions to modify the agent's behavior and persona during review phases. Evidence: File 'references/phase1-spec.md' instructs the agent to act as a 'hyper-critical spec reviewer,' and 'references/phase3-4-adversarial.md' mandates the persona of 'Sarcasmotron,' a reviewer with 'zero patience' and 'zero tolerance' who must provide no positive feedback.
[EXTERNAL_DOWNLOADS]: The skill requires downloading verification and testing tools from official registries. Evidence: 'references/phase5-6-hardening.md' includes commands to install '@stryker-mutator/core', 'mutmut', 'cargo-mutants', and 'Semgrep'. These are well-known industry-standard tools.
[COMMAND_EXECUTION]: The methodology involves executing shell commands for package management and running test suites. Evidence: The pipeline executes 'npm install', 'pip install', 'cargo install', and executes testing frameworks like Stryker and Semgrep through the command line.
[INDIRECT_PROMPT_INJECTION]: The skill's workflow creates an attack surface for instructions embedded in analyzed data to influence agent behavior. Ingestion points: Project files including 'vsdd/spec.md' and application source code. Boundary markers: Simple text labels (e.g., 'SPEC:') are used without explicit 'ignore instructions' directives. Capability inventory: Extensive capabilities including file system writes, package installations, and execution of test suites. Sanitization: No filtering is performed on the content of analyzed files before processing.
[DYNAMIC_EXECUTION]: The skill generates and executes code through its TDD and mutation testing phases. Evidence: The agent is instructed to write implementation code and immediately execute it through various test runners to verify behavior.

vsdd