The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data, including prompt templates from local JSON files and datasets from the Langfuse API, which are then interpolated into LLM completion requests.
Ingestion points: The scripts/deploy-prompts.mjs script reads from src/prompts/*.json, and tests/ai/experiment-gate.test.ts ingests data from the intent-classification-test dataset.
Boundary markers: No specific delimiters or boundary markers are implemented in the OpenAI completion calls to separate instruction from data.
Capability inventory: The skill utilizes file system access (fs), network operations (OpenAI and Langfuse SDKs), and shell execution capabilities via the allowed Bash(gh:*) tool.
Sanitization: There is no evidence of sanitization or validation of the template content or dataset records before they are processed by the LLM.
[EXTERNAL_DOWNLOADS]: The skill configuration and instructions involve fetching dependencies and tools from established services during the CI process.
Details: The GitHub Actions workflow utilizes official actions such as actions/checkout@v4 and actions/setup-node@v4 to prepare the execution environment.
Details: Dependencies are managed and executed using standard Node.js tools, including npm ci and npx vitest, which fetch packages from the public NPM registry.

langfuse-ci-integration