Agentic Development Platform Engineer

We’re hiring two engineers to build, run, and continuously improve the agentic development lifecycle (ADLC) platform underneath that work. The platform includes the shared harness and skill library, the evaluation and observability pipelines, the security controls, and the automation that turns every change into SOC2 Type 2 and FedRAMP High evidence. Our goal is to enable engineers to supervise coordinated agents delivering multiple product features in parallel, with clear human accountability for architecture, quality, and final acceptance.

This is a software-engineering role with a platform specialty. You will deliver real product changes with pods and use that work to validate and improve the shared development system.

WHAT YOU WILL DO

Build the agentic development flywheel

• Design orchestration that decomposes approved specifications into tasks, assigns specialized agents, and coordinates dependencies and handoffs across implementation, testing, review, and documentation.

• Enable concurrent feature work across repositories using isolated branches, worktrees, and environments. Manage shared dependencies, conflicts, concurrency limits, and integration through reviewed pull requests.

• Create human checkpoints and escalation paths for architecture, product decisions, sensitive actions, and final acceptance. Product teams retain ownership of their systems and domain decisions.

Build and run the platform

• Design, build, and operate the shared agentic engineering harness used by every pod. This covers Claude Code configuration, system prompts and rule files, hooks, subagents, MCP server connections, and OpenSpec templates and workflows. Keep models, coding agents, and orchestration tools replaceable through stable interfaces.

• Own the shared skill library as production code: versioned, pinned, reviewed in PRs, tested in CI, and distributed to every repository.

• Build context systems that give each agent accurate, task-relevant repository knowledge, architecture, domain rules, specifications, and prior decisions within appropriate context and access limits.

• Build and maintain evaluation pipelines. For skills, that means trigger accuracy, execution, regression, and token-budget evals. For agent work, it means trajectory-aware and final-output evals. Run fast checks in CI with minimum pass thresholds and broader scheduled evaluations for deeper regression, quality, and cost testing.

• Instrument agent runs with end-to-end tracing of sessions, agent trajectories, tool calls, context, outputs, failures, retries, and handoffs. Add cost and token metering, and sample so that failures and runaway loops are always kept. Apply appropriate redaction, access controls, and retention to traces.

• Keep the platform highly available across environments. Debug agent failures across the stack, including context overflow, tool misrouting, and self-repair loops that burn budget without progress. Build bounded retries, checkpointing, resumability, cancellation, and idempotent tool actions so partially completed work can recover safely.

Continuously improve the tooling

• Join pods for targeted product initiatives and deliver selected changes from specification through testing, deployment, and operation. Use these outcomes to validate agent workflows and identify reusable improvements.

• Use eval results, traces, and delivery metrics to find where the harness is failing engineers, and then fix it. Turn recurring failures into regression cases and validated improvements into shared context, skills, and workflows.

• Maintain the skill library over time. Harvest proven workflows from successful traces into new skills, tune skill descriptions for trigger accuracy, and retire skills that don’t earn their context cost.

• Run bounded experiments on harness changes such as models, prompts, skills, and tools. Ship only what measurably improves outcomes without regressing the existing suite.

• Partner with pods to capture domain knowledge as skills. Review and promote pod-contributed skills through the same eval and deployment gates as platform-owned ones.

• Build shared default workflows for AI tooling across repositories, including evaluating tools and making procurement recommendations. Give pods room to experiment and promote proven patterns when they reduce repeated work, cost, or risk.

Secure and govern agent activity

• Enforce least privilege for agents through dedicated agent identities, scoped short-lived credentials, and permission policies per agent and environment.

• Harden the tool layer. Vet and allowlist MCP servers, defend against prompt injection and spoofed tools, and add deterministic hooks and policy checks before high-risk actions.

• Integrate SAST, SCA, secret scanning, and security review tuned to the failure modes of AI-generated code into CI.

• Maintain context hygiene so that PII, secrets, and other sensitive data stay out of agent context and outputs.

Meet compliance and reporting requirements

• Build and maintain the automation behind our SDLC gates in GitLab. This includes validating spec approval (named approver, spec commit SHA, security and privacy fields) and creating Stories automatically from approved specs.

• Keep the Initiative Epic Story PR chain a complete, queryable audit trail. Partner with Security and Compliance to produce SOC2 Type 2 and FedRAMP High evidence on demand.

• Apply the same change-management controls to harness artifacts (prompts, rule files, skills, agent configurations) that apply to application code.

• Report on engineering outcomes and pipeline health: accepted changes per engineer, delivery cycle time, reviewer effort, rework, escaped defects, reliability, cost per accepted change, gate cycle time, review queue depth, eval pass rates, and DORA metrics.

• Document the platform with runbooks, onboarding guides, and standards that engineers can follow without asking.

• 3+ years (mid-level) or 5+ years (senior) of product software-engineering experience, with additional depth in developer platforms, DevOps, SRE, developer productivity, or test infrastructure engineering AT4

• A bachelor’s degree, technical certification, or equivalent work experience

• Daily, hands-on use of agentic coding tools (Claude Code preferred), with working knowledge of harness components such as skills, commands, hooks, subagents, and MCP, and demonstrated ability to turn agentic workflows into repeatable, tested engineering capabilities

• Strong coding ability in TypeScript/JavaScript and/or Python, experience delivering and operating production software, and comfort working in unfamiliar application code (our backend is Elixir)

• Experience building CI/CD pipelines and managing infrastructure as code (Terraform, CloudFormation)

• Experience designing test or evaluation suites, plus a working understanding of LLM evaluation, including LLM-as-judge and its calibration pitfalls

• Experience with observability tooling such as Prometheus, Grafana, and OpenTelemetry

• A solid grasp of cloud security fundamentals: IAM, least privilege, secrets management, and supply-chain scanning

• Experience working in a regulated environment (SOC2, FedRAMP, CJIS, HIPAA, PCI, or similar) and producing audit evidence

• Experience automating GitLab through its APIs, webhooks, or workflow rules

• Clear technical writing, because you’ll author specs, skills, and standards that both engineers and agents follow

• Strong plus: building or maintaining MCP servers; spec-driven development or OpenSpec experience; FedRAMP High or CJIS experience; familiarity with agent threat models such as prompt injection, confused deputy, and denial of wallet; Docker and Kubernetes; SSH and terminal fluency; distributed workflow systems, multi-agent orchestration, or repository-specific context systems

• Judgment over output: you measure success by whether engineers can oversee more accepted, high-quality changes with less rework and reviewer burden, and whether they trust the tools

• Integrity, Honesty, and Forthrightness

• Dedication to the mission of Public Safety

hackajob is collaborating with Leo Technologies to connect them with exceptional professionals for this role.