The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill utilizes a custom Javascript engine located in 'references/qa-execute-snippets.mjs' to dynamically parse, classify, and execute shell code snippets found in markdown files at runtime.
[COMMAND_EXECUTION]: The skill is designed to invoke various system utilities and developer tools, including 'git', 'shasum', 'sed', 'grep', and ecosystem-specific runners such as 'pytest', 'npx tsc', and 'cargo check' to audit the integrity of the filesystem and artifact correctness.
[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection because it ingests and executes commands derived from untrusted markdown content produced or modified during the agent's session. * Ingestion points: Reads artifact descriptions, requirement ledgers, and repository markdown files during the audit process. * Boundary markers: Implements an internal 'SAFE_COMMANDS' allow-list and 'DENY_PATTERNS' to filter shell execution. * Capability inventory: Spawns shell processes via 'child_process.spawnSync' and performs recursive file inspections. * Sanitization: Features a sophisticated Bash lexer and tokenizer to verify command names and prevent the use of escaped or obfuscated dangerous commands.

double-check