The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various shell commands for project environment detection, type checking, and testing.
Evidence: SKILL.md contains commands such as bunx tsc --version, node -e, and bun run typecheck || bunx tsc --noEmit.
Evidence: scripts/ts_diagnostic.py uses subprocess.run(cmd, shell=True) to execute hardcoded diagnostic commands like npx tsc --version and grep.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project data (e.g., package.json, tsconfig.json, and source code) which could theoretically contain malicious instructions targeting the agent, although the tools used (tsc, grep, node) are standard for this domain.
Ingestion points: The agent reads package.json, tsconfig.json, and all files in the src/ directory.
Boundary markers: None identified.
Capability inventory: The skill can execute shell commands via the agent's shell tool and the ts_diagnostic.py script.
Sanitization: No explicit sanitization or filtering of file content is performed before passing it to development tools or interpreting its structure.
[SAFE]: All external URLs point to well-known documentation sites (e.g., TypeScript Handbook, Microsoft GitHub, Vitest, Egghead.io) or the author's own domain (shipshit.dev). The use of subprocess.run(shell=True) in the diagnostic script is restricted to static, hardcoded command strings.

typescript-expert