The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes grep and sed via the Bash shell to query the local references/coverage-data.md file. The commands are defined as static templates in the instructions to ensure they are used only for intended data retrieval.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied CVE and AVD IDs as query parameters. The instructions mandate normalization and specific character anchoring (e.g., using pipe delimiters in regex) to prevent exploitation of the search logic or incorrect data matching.
[DATA_EXFILTRATION]: The skill instructions strictly prohibit network access and the use of external web tools. Furthermore, the agent is directed to use targeted shell commands rather than loading the entire data file into memory, mitigating risks related to context poisoning or unauthorized data exposure.

alibabacloud-security-vuln-coverage-check