The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads its constituent files and third-party research scripts from the author's GitHub repository and a Huawei Cloud OBS bucket (gts-download.obs.myhuaweicloud.com). These sources are associated with the Gangtise service and are necessary for the skill's stated purpose as a suite installer.
[COMMAND_EXECUTION]: The installation, credential configuration, and diagnostic processes rely on shell scripts (install_gangtise.sh, configure_auth.sh, diagnose.sh) that perform file system modifications, including creating directories, symlinking configuration files, and enforcing strict file permissions (e.g., chmod 600 on sensitive files).
[DYNAMIC_EXECUTION]: SKILL.md contains a Python script that the agent is instructed to execute to download and write the skill's components to the local filesystem. This involves programmatic file creation based on remote content fetched via the GitHub API.
[INDIRECT_PROMPT_INJECTION]: The skill suite provides an interface to query and process large volumes of external data, including research reports, analyst opinions, and web search results, creating an attack surface for instructions embedded in data.
Ingestion points: Gangtise RAG API results, knowledge base search content, and file summaries retrieved via the kb.py and report.py scripts.
Boundary markers: No explicit boundary markers or "ignore instructions" delimiters are implemented in the provided scripts to wrap ingested content before agent processing.
Capability inventory: The skill suite possesses capabilities for shell command execution, network operations, and file system writes across its management and research scripts.
Sanitization: While classify_response.py cleans data for diagnostic logs, the primary research data processed by the agent lacks structural sanitization or filtering to prevent instruction adherence.

gangtise-copilot