The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing several widely-used development and security tools including mypy, pytest-cov, semgrep, gitleaks, osv-scanner, lighthouse, size-limit, axe-core, dependency-cruiser, and Stryker. These downloads are facilitated through official package managers (npm, pip, brew) targeting well-known services and repositories, which is considered standard practice for project environment setup.\n- [COMMAND_EXECUTION]: The skill provides example commands for executing the aforementioned quality and security tools (e.g., tsc --noEmit, eslint ., gitleaks detect). These commands are necessary for the skill's primary purpose of enforcing project constraints and are executed locally within the project context. The skill also explicitly promotes security best practices by including a specific warning to use the --redact flag with gitleaks to prevent sensitive data from entering logs.\n- [DYNAMIC_EXECUTION]: The provided reference script floor-guard.mjs uses node:child_process's execFileSync to invoke local git commands. This is used to calculate the diff between the current working tree and the merge base to perform diff-scoped validation of code quality. The script operates locally and does not perform network operations or access sensitive system paths outside of the repository.\n- [INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface where it reads and processes the output of git diff and the content of CONSTRAINTS.md. This represents a potential indirect prompt injection surface if malicious content were embedded in those files; however, the script's logic is limited to identifying specific patterns (e.g., suppression comments, lowered thresholds) and does not interpolate this data into executable code or downstream agent instructions in a way that creates a high-risk capability chain.

constraint-driven-development