BLACKSHIELD

Guide public

Deploy the Security Agent

Run BlackShield's governed Go security agent inside your environment with tenant-scoped assignments, MCP context, AI Gateway controls, and approval gates around every mutation. Audience: Security engineers, platform engineers, and security operations teams. Temps moyen de mise en place: 5 minutes.

quickstart

Use this if

Run BlackShield's governed Go security agent inside your environment with tenant-scoped assignments, MCP context, AI Gateway controls, and approval gates around every mutation.

Audience
Security engineers, platform engineers, and security operations teams
Typical time
5 minutes

Avant de commencer

  • Confirm the tenant is on a plan with Security Agents enabled and `/agents` is visible to the admin team.
  • Connect at least one provider in AI Gateway before booting the runtime so the planner has a valid governed inference path.
  • Decide where the agent will run and what local command surface it truly needs before mounting Docker sockets, repos, or host paths.

Fast path

Copy a working starter, run it in your environment, then come back here for the deeper rollout details.

Get the source bundle

Download the exact source files referenced on this page, or run the one-command installer to write them locally before following the deployment steps.

Security agent Docker Compose bundle

Creates `deploy/security-agent/` with a Compose file, a ready-to-edit `.env.security-agent`, and a clean example copy so the agent can bootstrap with the documented runtime settings.

deploy/security-agent/
bash
BLACKSHIELD_SECURITY_AGENT_IMAGE=public.ecr.aws/blackshield-security/security-agent:1.0.0 \
BLACKSHIELD_SITE_URL=https://blackshield.chaplau.com \
BLACKSHIELD_API_URL=https://api.blackshield.chaplau.com \
bash <(curl -fsSL https://blackshield.chaplau.com/source-bundles/security-agent-docker.sh)
cd deploy/security-agent

Exécuter

Docker Compose bootstrap

yaml
# Bootstrap the local runtime bundle first
bash <(curl -fsSL https://blackshield.chaplau.com/source-bundles/security-agent-docker.sh)

# Edit deploy/security-agent/.env.security-agent and replace BLACKSHIELD_API_KEY
cd deploy/security-agent
docker compose -f docker-compose.security-agent.yml up -d
docker compose -f docker-compose.security-agent.yml logs -f

One-shot runtime launch

bash
docker run --rm \
  -e BLACKSHIELD_API_URL=https://api.blackshield.chaplau.com \
  -e BLACKSHIELD_API_KEY=sagent_xxxx \
  -e SECURITY_AGENT_MODEL=gpt-4o-mini \
  -e SECURITY_AGENT_PROVIDER=openai \
  -e SECURITY_AGENT_MAX_STEPS=8 \
  -e SECURITY_AGENT_COMMAND_TIMEOUT_SECONDS=120 \
  -v /var/run/docker.sock:/var/run/docker.sock \
  -v "$(pwd):/workspace" \
  -w /workspace \
  public.ecr.aws/blackshield-security/security-agent:1.0.0

Assignment payload example

json
{
  "assignment_type": "security_review_support",
  "title": "Review internet-facing payment service",
  "instructions": "Summarize open control gaps and propose review answers with evidence.",
  "scope": {
    "review_id": "2c79f0df-42b8-4d7f-9d28-7a7f4c5d6b01"
  }
}

Understand and customize

Use the guided steps below when you want to tailor the rollout, validate ownership, or expand the deployment safely.

Étape 1

Create an installation and save the bootstrap key

Use the Agents console so the runtime gets a dedicated installation identity instead of reusing a human API key.

  • Open `/agents` and create a Security Agent installation for the target tenant or environment.
  • Copy the bootstrap key shown after installation creation and store it in your secret manager.
  • Review the effective policy profile, allowed providers, and assignment types before deployment.
  • Confirm AI Gateway already has at least one approved provider connection for the tenant.

What success looks like

Confirm AI Gateway already has at least one approved provider connection for the tenant.

Étape 2

Bootstrap the runtime in the target environment

Use the Docker Compose bundle for the fastest start, then mount only the local paths and executables the agent actually needs.

  • Use the source bundle on this page to create `deploy/security-agent/` with a compose file, a ready-to-edit `.env.security-agent`, and a clean `.example` reference copy.
  • Replace `BLACKSHIELD_API_KEY` in `.env.security-agent`, then tune `SECURITY_AGENT_MODEL`, `SECURITY_AGENT_MAX_STEPS`, and command timeout values only if needed.
  • Run the container with the minimum local filesystem or Docker socket access your workflow requires.
  • Keep the runtime close to the tenant assets it needs to inspect so read-only local commands stay low-latency and auditable.

What success looks like

Keep the runtime close to the tenant assets it needs to inspect so read-only local commands stay low-latency and auditable.

Étape 3

Queue a blue-team assignment and review approvals

Start with read-first work such as finding triage, remediation follow-up, or security review support, then approve only the actions you want executed.

  • Create an assignment in `/agents` for `finding_triage`, `remediation_followup`, or `security_review_support`.
  • Watch the run timeline, planner observations, and uploaded artifacts to confirm the agent has enough context.
  • Use the approval queue for any mutating local command or product-side remediation or review update.
  • Download the final report and fold it into your incident, remediation, or review workflow.

What success looks like

Download the final report and fold it into your incident, remediation, or review workflow.

What success looks like

  • The agent claims an assignment from `/api/v1/security-agents/agent/assignments/claim` and starts a run successfully.
  • Planner events and artifacts appear in the Agents console for the run without exposing tenant provider secrets to the runtime.
  • A mutating action pauses for approval and resumes correctly after approval is granted.
Deploy the Security Agent | BlackShield Docs