The Agent Skills Directory

[COMMAND_EXECUTION]: The skill directs the agent to execute python tests/run_tests.py, allowing for arbitrary code execution from the local filesystem.
[DATA_EXFILTRATION]: The skill accesses ~/.claude/agents/, a hidden directory containing agent configuration, which could lead to exposure of internal setup data.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its ingestion of multiple project files. Ingestion points: files in agent-improvements/ and mel_wiki/. Boundary markers: None present in the instructions. Capability inventory: shell command execution and deep file system access. Sanitization: no validation or filtering is applied to the ingested content.

grade-system