The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates by reading and processing work item descriptions, requirements, and comments from the Yunxiao platform. This creates an attack surface where malicious instructions embedded in these external data sources could attempt to influence the agent's behavior.
Ingestion points: Data is retrieved from the Projex API via scripts such as scripts/yunxiao_cli_allocate_task.py and scripts/yunxiao_cli_bug_batch.py using official CLI commands.
Boundary markers: The skill employs HTML comment markers (e.g., ) and structured JSON to separate machine-readable data from user-authored text.
Capability inventory: The skill can execute shell commands via the aliyun CLI, write to work items, and perform network requests to the Yunxiao API.
Sanitization: The skill implements a scrub function in scripts/yunxiao_cli_runtime.py to redact tokens and secrets from its output logs and terminal displays.
[DYNAMIC_EXECUTION]: The script scripts/ensure-daily-skill-update.mjs executes npx skills@latest update to maintain the skill suite. This involves downloading and running the latest version of an external package from the public npm registry at runtime, which poses a supply chain risk if the package is untrusted or compromised.
[COMMAND_EXECUTION]: Numerous scripts, including scripts/yunxiao_cli_runtime.py and scripts/yunxiao_cli_bug_batch.py, use subprocess.run to call the aliyun devops CLI. These calls are constructed using argument lists, which is a standard and generally safe practice, but they do perform state-changing operations on the remote platform.
[EXTERNAL_DOWNLOADS]: The skill fetches data from the Yunxiao API (https://openapi-rdc.aliyuncs.com) and downloads document content from URIs specified in work item manifests using urllib.request.urlopen. While it performs integrity checks using SHA-256 hashes, this mechanism allows for automated retrieval of content from external domains.

yunxiao-development-delivery