The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local repository, including source code (e.g., bot.py) and existing JSON test specifications, to automatically generate expected_outcome statements and validation criteria for an LLM judge. While the skill includes a dedicated linter (scripts/lint_suite.py) to validate tags and speaker labels, the interpolation of repository-derived phrases into the judge's prompt creates a surface for indirect prompt injection.
Ingestion points: Repository source files (scanned via grep and git) and cekura.tests.json specification files.
Boundary markers: The skill utilizes a structured JSON schema (https://docs.cekura.ai/schemas/test-suite/v1.json) to delimit fields, but the expected_outcome field contains natural language instructions for the judge.
Capability inventory: The skill executes local Python scripts, performs repository-wide file reads, and makes network requests to the vendor's API.
Sanitization: The linter script performs basic checks for tag syntax and restricted speaker labels to prevent transcript confusion.
[EXTERNAL_DOWNLOADS]: The CI/CD integration templates provided in references/ci-wiring.md (specifically for GitLab CI) include instructions to download an external Python runner script from the vendor's repository (https://raw.githubusercontent.com/cekura-ai/cekura-github-actions/v1.3.0/run-suite/run_suite.py) during the build process.
[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands such as grep and git ls-files to discover repository structure and identify existing test suites. Additionally, it executes local Python scripts (scripts/lint_suite.py and scripts/run_suite.py) which interact with the file system and network to validate and upload test results.

cekura-infra-test-suite