The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to execute npx debug-agent, which downloads and runs a package from the public npm registry. The use of an unversioned and non-namespaced package is a security concern as it could lead to the execution of malicious code if the package name is squatted or the registry is compromised.
[DATA_EXFILTRATION]: The instructions require the agent to write runtime logs to /tmp/debug-agent/. Files stored in /tmp are often accessible to all users on a multi-user system, potentially exposing internal application state and logic.
[DYNAMIC_EXECUTION]: The skill workflow involves instrumenting the target code by dynamically inserting new executable logic (such as network requests or file writes) into existing source files.
[INDIRECT_PROMPT_INJECTION]: The agent is tasked with reading and analyzing runtime logs produced by the instrumented code. If the code under test processes untrusted external data that appears in the logs, it creates a vector for indirect prompt injection where malicious instructions could influence the agent's debugging and fixing logic. * Ingestion points: The agent reads the log file located at the dynamically determined logPath in SKILL.md. * Boundary markers: None identified; the agent parses raw NDJSON lines. * Capability inventory: The agent has the ability to execute shell commands (npx), modify source code (instrumentation), and delete files. * Sanitization: There are no instructions to sanitize or validate the log content before the agent evaluates it to make code changes.

debug-agent