The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the user's project environment (such as package.json, tsconfig.json, and TypeScript source files) which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: The skill reads package.json and tsconfig.json via both Python scripts and shell commands. It also recursively scans the src/ directory for type usage patterns in scripts/ts_diagnostic.py and SKILL.md.
Capability inventory: The skill can execute shell commands (npx, grep, node -e) and has access to a Python diagnostic script that uses subprocess.run to execute commands.
Boundary markers: Absent. The instructions do not define clear delimiters or warnings to ignore instructions found within processed project files.
Sanitization: Absent. Data read from external files is printed directly to the agent's context or processed by shell utilities without escaping or validation.
[COMMAND_EXECUTION]: The skill uses various shell commands to detect the environment and run diagnostics.
Evidence: SKILL.md contains instructions for the agent to run complex one-liners like node -e for tooling detection and npx tsc --noEmit for validation. scripts/ts_diagnostic.py executes commands like npx tsc --version, node -v, and several grep operations.
[DYNAMIC_EXECUTION]: The skill utilizes dynamic code execution to perform its tasks.
Evidence: In SKILL.md, a node -e command is used to execute a JavaScript snippet that parses package.json to detect installed tools. In scripts/ts_diagnostic.py, shell commands are executed using subprocess.run with shell=True in the run_cmd function.

typescript-expert