The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Automated security scans identified references/coverage.md as malicious (detected as FileRepMalware). This indicates the file contains patterns or signatures associated with malware, which overrides the safety of the visible documentation text.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a vulnerability surface by ingesting untrusted data from the local environment.\n
Ingestion points: The skill parses SF, LF, LH, and DA records from the coverage/lcov.info file, as described in SKILL.md and references/coverage.md.\n
Boundary markers: No boundary markers or delimiters are defined to isolate external data from agent instructions during the parsing process.\n
Capability inventory: The skill uses Bash, Edit, Write, and mcp__very-good-cli__test tools, which could be exploited if malicious instructions are embedded in the coverage data.\n
Sanitization: The instructions do not specify any validation, escaping, or filtering of the content read from the lcov.info file before it is processed.\n- [COMMAND_EXECUTION]: The skill uses Bash to parse the contents of coverage/lcov.info. Processing external file content via shell commands without explicit sanitization of file paths or record values introduces a risk of command injection, particularly if filenames or metadata within the project are crafted maliciously.\n- [EXTERNAL_DOWNLOADS]: The skill provides instructions for the user to install external tooling (very_good_cli) via dart pub global activate very_good_cli if the necessary environment hooks are not detected.

green-gate