The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/find-polluter.mjs contains a command injection vulnerability. The quote function fails to sanitize shell metacharacters such as backticks or dollar signs when wrapping filenames in double quotes. When executed via spawnSync with shell: true, these characters allow for arbitrary command execution through shell substitution by the system shell.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data, creating an attack surface for indirect prompt injection.
Ingestion points: The skill reads error messages, stack traces, HAR files, and log dumps provided by the user or external systems as described in SKILL.md.
Boundary markers: No explicit delimiters or instructions to ignore embedded instructions are provided for handling diagnostic data.
Capability inventory: The agent has the ability to execute shell commands via the provided scripts.
Sanitization: Instructions focus on masking credentials but do not include measures to sanitize content to prevent the agent from following malicious instructions hidden in processed data.
[DYNAMIC_EXECUTION]: The find-polluter.mjs script dynamically builds shell commands by interpolating file lists into command templates and executes them using spawnSync with the system shell enabled.

diagnosing-bugs