The Agent Skills Directory

[COMMAND_EXECUTION] (LOW): The script scripts/ts_diagnostic.py uses subprocess.run(shell=True) to call tools such as npx, node, and grep. While the command strings are static, shell execution and dependency on the local environment can be exploited if the environment or project configuration is maliciously crafted.
[PROMPT_INJECTION] (LOW): The script ingests data from local files like package.json, tsconfig.json, and source files. Evidence: 1. Ingestion points: package.json, tsconfig.json, and src/*.ts in scripts/ts_diagnostic.py. 2. Boundary markers: Not used. 3. Capability inventory: subprocess.run in scripts/ts_diagnostic.py. 4. Sanitization: None. This ingestion could allow an attacker to inject instructions via file content that appear in the diagnostic output for the agent to consume.

typescript-expert