The Agent Skills Directory

[DATA_EXFILTRATION]: Telemetry data is sent to an external domain https://skillsop.topxtopx.com/api/quality/guest-report. This data includes the user's intent, user_input, output_result, and execution steps. While the SDK attempts to mask secrets and phone numbers, the transmission of interaction history to a non-whitelisted domain constitutes a data exposure risk. Additionally, the SDK reads local session databases from tools like Opencode, Hermes, and Codex to extract usage statistics.
[COMMAND_EXECUTION]: The skill executes shell commands generated during its technical research phase using bash -c. Although the script test-cli-commands.sh enforces an allowlist (hcloud, python3, curl, bash), it still executes AI-generated logic derived from user-specified requirements.
[DYNAMIC_EXECUTION]: The skill dynamically executes Python code and Node.js scripts. Specifically, skill_quality_sdk.py uses subprocess.run to execute a Node.js proxy for token retrieval and runs SQL queries against local SQLite databases. report-skill-created.sh also dynamically locates and executes a reporting script (report.mjs) from the local agent environment.
[INDIRECT_PROMPT_INJECTION]: As a code generator, the skill has an attack surface where user-provided requirements could lead to the generation of malicious skill instructions or commands. It attempts to mitigate this through Socratic questioning and strict YAML validation of the generated frontmatter.

huawei-cloud-skill-creator