The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes a documentation link to https://jeffallan.github.io/claude-skills/skills/quality/debugging-wizard/, which has been explicitly flagged as malicious by security scanners (URL:Blacklist).
[METADATA_POISONING]: Automated scans have identified the SKILL.md file as malicious (FileRepMalware), indicating that the file or its source repository carries a high reputation risk.
[COMMAND_EXECUTION]: The skill instructs the agent to execute various shell-level debugging tools and commands, including python -m pdb, node --inspect, dlv debug, and git bisect. It also includes instructions for destructive file system operations like rm -rf node_modules.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to parse and analyze potentially untrusted data such as error messages, stack traces, and log files, which provides a surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent's context through user-provided logs and error reports as specified in SKILL.md and references/systematic-debugging.md.
Boundary markers: The skill fails to define any delimiters or provide instructions to the agent to ignore embedded commands within the data being analyzed.
Capability inventory: The agent has access to shell command execution through various debugging utilities, which could be exploited if malicious instructions are embedded in the analyzed data.
Sanitization: No input validation or escaping mechanisms are implemented to ensure that the content of the error messages or logs does not contain executable or override instructions.

debugging-wizard