Public documentation

Environment Variables

CLI Target Variables

Variable	Required	Description
`AGENT_API_KEY`	optional	Example credential for your own HTTP agent target.

Built-In Attack Pack Target

Set exactly one:

Variable	Description
`ROLEPLAY_TARGET_URL`	HTTP target URL for `social-engineering-core`.
`ROLEPLAY_TARGET_COMMAND`	CLI target command for `social-engineering-core`.

Attacker And Judge Providers

Use these when you want adaptive attacker turns and semantic or hybrid transcript judging instead of the local mock smoke-test provider.

Variable	Required	Description
`ROLEPLAY_LLM_PROVIDER`	for real adaptive runs unless `--provider` is provided	Shared provider for attacker and judge.
`ROLEPLAY_ATTACKER_PROVIDER`	optional	Provider for adaptive attacker turns. Overrides `ROLEPLAY_LLM_PROVIDER`.
`ROLEPLAY_JUDGE_MODE`	for real runs unless `--judge` is provided	`rules`, `semantic`, or `hybrid`. Use `hybrid` for CI and serious real-agent tests.
`ROLEPLAY_JUDGE_PROVIDER`	for semantic/hybrid judging unless `--judge-provider` is provided	Provider for transcript judging. Overrides `ROLEPLAY_LLM_PROVIDER`.
`ROLEPLAY_LLM_MODEL`	optional	Shared model name.
`ROLEPLAY_ATTACKER_MODEL`	optional	Model for adaptive attacker turns.
`ROLEPLAY_JUDGE_MODEL`	optional	Model for judging.
`ROLEPLAY_OPENAI_API_KEY`	for `openai`	OpenAI API key.
`ROLEPLAY_ANTHROPIC_API_KEY`	for `anthropic`	Anthropic API key.
`ROLEPLAY_GOOGLE_API_KEY`	for `google`	Google Gemini API key.
`ROLEPLAY_LLM_API_KEY`	for `openai-compatible`	API key for an OpenAI-compatible provider.
`ROLEPLAY_LLM_BASE_URL`	for `openai-compatible`	Provider base URL, for example `https://api.example.com/v1`.

Provider identifiers are openai, anthropic, google, and openai-compatible. They are reference options, not defaults. Use --target mock --provider mock --judge rules only for install smoke tests.

Workbench Upload Variables

Variable	Required	Description
`ROLEPLAY_CLOUD_URL`	for uploads unless `--endpoint` is provided	workbench base URL.
`ROLEPLAY_PROJECT_ID`	for uploads unless `--project` is provided	workbench project ID.
`ROLEPLAY_API_KEY`	for uploads unless `--api-key` is provided	Project API key.
`ROLEPLAY_AGENT_NAME`	recommended for multi-agent projects	Target agent attribution.
`ROLEPLAY_ENVIRONMENT`	optional	Environment label for uploaded run.
`ROLEPLAY_BUILD_URL`	optional	CI build URL.

CI Detection

The CLI can infer build metadata from common CI variables.

Branch:

GITHUB_REF_NAME
BRANCH_NAME

Commit:

GITHUB_SHA
COMMIT_SHA

Build URL:

ROLEPLAY_BUILD_URL
GitHub Actions URL from GITHUB_SERVER_URL, GITHUB_REPOSITORY, and GITHUB_RUN_ID
CI_JOB_URL
CIRCLE_BUILD_URL
BUILD_URL