The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file tools/docsbot.py contains a hardcoded DOCSBOT_API_KEY (ba25f6f95c196d017919b550539a2cf863657d220345133aba40f86a3969d623) for documentation retrieval via the DocsBot AI service.
[DYNAMIC_EXECUTION]: tools/apply.py dynamically loads product-specific validation scripts from computed file paths using importlib.util.spec_from_file_location. Although the input is validated against a regular expression, this pattern executes code from paths determined at runtime.
[OBFUSCATION]: The signature generation library lib-generate-test-usersig-es.min.js is minified. Additionally, the core telemetry logic is provided as a pre-bundled file telemetry.cjs, which obscures the underlying code execution during reporting.
[COMMAND_EXECUTION]: Scripts in the runtime/ and tools/ directories utilize subprocess.Popen and subprocess.run to interact with system tools like adb and xcrun for device log collection, and node for executing the telemetry bundle.
[EXTERNAL_DOWNLOADS]: The skill fetches documentation and submits telemetry to external domains, including api.docsbot.ai, trtc.io, and ap-nanjing.cls.tencentcs.com. The latter two are vendor domains, but DocsBot is a third-party service.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: User queries are ingested by the root dispatcher in SKILL.md and passed directly to classification tools like tools/query_classifier.py and tools/search.py.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present when interpolating user queries into tool inputs.
Capability inventory: Subprocess execution for reporting and validation, local file system writes for session state management, and network requests for documentation retrieval.
Sanitization: A centralized redaction module (runtime/redact.js) implements 14 rules to scrub sensitive patterns from reported text before it is sent to the backend.

trtc