## Summary You can now discover and invoke GitLab Duo Agent Platform (DAP) Agents directly from external AI-native tools like Cursor, VS Code, and terminal-based workflows via the GitLab MCP server, without ever having to leave your development environment. This brings the AI agents your team has built and standardized on inside GitLab to wherever you actually work, so you get the full power of GitLab's AI orchestration and governance no matter which tool you're coding in. ## Test script Test scenario for invoking DAP foundational + custom agents from a third-party MCP client. Persona: a developer working in **Claude Code** who never leaves the IDE but reaches into GitLab's agents as callable MCP units. The beats chain into one SDLC storyline: **plan → research → human codes → review (three lenses) → diagnose CI**. The human writes the code; the agents wrap around the coding task. Support agentic mode 2 (hybrid - not fully agentic). > [!NOTE] > Customer context is **healthcare / HIPAA** so the custom agent is distinct from the security agent. #### Flow overview | # | Agent | Dev's situation (in Claude Code) | What it stresses | |---|-------|-----------------------------------|-------------------| | 1 | Planner | "Turn this feature idea into a plan" | Long-running + **write** ops + HITL approval | | 2 | Duo Dev (research) | "Before I code, how do we do Okta SSO today?" | Read across repo + Knowledge Graph; context flows back into the IDE | | — | *Dev codes the change in Claude Code, opens !88* | *(human in the loop — the point of the demo)* | — | | 3 | Duo Code Review | "Review my MR !88" | Reads human diff, returns structured findings; async | | 4 | CI Expert | "Pipeline on !88 is red — why?" | Server-side job logs the local agent can't see | | 5 | Security Agent | "Any vulns in !88?" | Cross-ref working tree + GitLab SAST / dep-scan | | 6 | Custom: HIPAA/PHI Reviewer | "Check !88 against our PHI controls" | Customer-authored agent + discovery | **Three reviews, one MR.** Beats 3, 5, 6 all review !88 — run them general → security → compliance and say the framing out loud so it reads as *layered governance*, not repetition: - Duo Code Review → code quality & correctness - Security Agent → vulnerabilities - HIPAA Reviewer → compliance controls --- ### Beat-by-beat test cases #### 1. Planner — write ops + HITL - **Prompt:** *"Use the GitLab Planner to break 'patient-portal SSO with Okta' into an epic with issues and a milestone."* - **Illustrative call:** `invoke_agent(agent="planner", input="patient-portal SSO with Okta", project=<id>)` - **Expected:** proposed epic + child issues + milestone returned for approval, *then* created in GitLab on confirm. | Check | Pass | |-------|------| | Plan returned as a proposal before anything is written | | | Writes gated behind explicit dev approval (not fire-and-forget) | | | Created epic/issues/milestone actually appear in GitLab | | #### 2. Duo Dev (research mode) — context back to the IDE - **Prompt:** *"Before I write this, ask Duo to research how our codebase handles Okta/OIDC today and what SSO patterns already exist."* - **Illustrative call:** `invoke_agent(agent="duo_developer", mode="research", input="existing Okta/OIDC + SSO patterns in this repo")` - **Expected:** synthesized findings — relevant files, existing auth utilities, prior art — returned as text the dev reads *before* coding. | Check | Pass | |-------|------| | Findings reference real files / utilities in the repo | | | Output is usable context Claude Code can act on (not a dead-end blob) | | | Draws on Knowledge Graph / cross-repo context the local agent lacks | | #### — Dev codes the change and opens MR !88 Human writes the feature in Claude Code using the plan (1) and research (2), pushes, opens **!88**. No agent call — this is the point of the demo. #### 3. Duo Code Review — reviewing a human-authored diff - **Prompt:** *"Run Duo Code Review on !88."* - **Illustrative call:** `invoke_agent(agent="duo_code_review", mr=88)` - **Expected:** structured review comments on the human diff — quality, correctness, maintainability. | Check | Pass | |-------|------| | Comments are specific to the actual diff (not generic) | | | Long-running call returns cleanly (handle/poll or stream, not a timeout) | | | Findings distinguishable from security/compliance lenses below | | #### 4. CI Expert — server-side context the local agent can't see - **Prompt:** *"The pipeline on !88 is failing. Ask the CI Expert what's wrong."* - **Illustrative call:** `invoke_agent(agent="ci_expert", mr=88)` - **Expected:** reads job logs + `.gitlab-ci.yml` server-side, returns root cause + suggested fix. | Check | Pass | |-------|------| | Diagnosis cites actual failing job / log lines | | | Uses pipeline data the local IDE does not have | | | Suggested fix is actionable in the repo | | #### 5. Security Agent — cross-reference diff + GitLab scan data - **Prompt:** *"Before I request review on !88, have the Security Analyst check it."* - **Illustrative call:** `invoke_agent(agent="security_analyst", mr=88)` - **Expected:** correlates the diff with GitLab SAST / dependency-scanning findings, returns prioritized issues. | Check | Pass | |-------|------| | Pulls GitLab-side SAST / dep-scan results | ☐ | | Prioritizes findings (not a flat dump) | ☐ | | Clearly a security lens, distinct from Duo Code Review | ☐ | #### 6. Custom HIPAA/PHI Reviewer — custom agent + discovery - **Prompt:** *"Run our PHI Data-Handling Reviewer on !88."* - **Illustrative call:** `invoke_agent(agent="highmark/phi-reviewer", mr=88)` - **Expected:** checks the change against customer-authored rules (no PHI in logs, encryption-at-rest for patient fields, audit-log coverage), returns pass/flag with citations to their control docs. | Check | Pass | |-------|------| | Applies customer-authored rules, not generic security checks | | | Cites the customer's own control docs | ☐ | | Custom agent is **discoverable** via `list_agents` (not just by hardcoded name) | | --- ### Hard-parts checklist Run these deliberately across the beats above to try to break it. | Dimension | How to test | Pass | |-----------|-------------|------| | **Long-running** | Does Duo Code Review (3) return a handle + let you poll, or hang/timeout? | | | **HITL** | Does Planner (1) gate writes behind approval, or fire-and-forget? | | | **Identity / permissions** | Run a beat as a dev who **lacks** write access — confirm it's denied, and confirm *whose* creds run the flow (dev PAT scope vs service identity). | | | **Trace continuity** | Does the trace survive the MCP boundary, or go dark once it leaves Claude Code? | | | **Discovery** | Can the client enumerate agents (`list_agents`), or must names be hardcoded? | | | **Error shape** | Kill a pipeline mid-diagnosis (4) — clean MCP error or a wall of stack trace? | | --- ### Notes for the team - Record which beats **blocked / timed out / returned a handle** — the async model for long-running flows (esp. Duo Code Review) is the biggest open design question. - Capture any beat where the trace goes dark at the MCP boundary. - Note whether custom-agent discovery worked without hardcoding names.