LbAgents 0.18.11

LangGraph-based agent workflows for LHCb Analysis Productions

LbAgents

LangGraph-based agent workflows plugged into LbAPI.

This package contains pure agent logic — graph definitions, prompt templates, state schemas, tools, and a Pydantic workflow registry. It has no infrastructure dependencies (no database, no HTTP server, no Celery). MCP tools are discovered at runtime and passed to graph factories by the host application (LbAPI).

Architecture

graph LR
    subgraph LbAgents ["LbAgents (this package)"]
        Registry[Workflow Registry]
        Factories[Graph Factories]
        Schemas[State Schemas]
        Prompts[Prompt Templates]
        Tools[GitLab Tools]
    end

    subgraph LbAPI ["LbAPI (host)"]
        Router["/agent/* router"]
        Celery[Celery tasks]
        MCP[MCP tool discovery]
        LLM[KServe LLM clients]
        PG[AsyncPostgresSaver]
        LF[Langfuse observability]
    end

    Router -->|imports| Registry
    Celery -->|builds graph from| Factories
    MCP -->|passes tools to| Factories
    LLM -->|passes LLMs to| Factories
    PG -->|compiles with| Factories

Key design decisions

• Factory pattern: Graph factories return an uncompiled StateGraph. The caller (LbAPI) compiles it with a checkpointer.
• Tools as parameters: Factories accept a tools list — no hardcoded tool definitions. LbAPI discovers MCP tools at runtime and combines them with GitLab tools from this package.
• Self-registration: Importing LbAgents populates WORKFLOW_REGISTRY with all available workflows. Adding a new graph to this package makes it available in LbAPI with zero LbAPI code changes.
• Write-tool interrupts: Tools that perform write operations (posting reviews, creating issues) are gated by human-in-the-loop interrupts. The pending tool call is shown to the reviewer before execution.

Workflows

Name	Description	Trigger	Delivery	Human Review
ap_debug_ci_failure	Single-agent CI failure diagnosis and MR review	Internal	GitLab MR	Before posting review
ap_debug_ci_failure_multiagent	Multi-agent CI failure diagnosis with parallel investigation	Internal	GitLab MR	Before posting review
ap_debug_stuck_productions	Diagnose stuck productions (MaxReset, stalled files)	Internal	GitLab Issue	Before write actions
mattermost_debug_user_request	Answer user support questions via Mattermost	Webhook	Mattermost	Always (admin verifies)
nightly_failure_summary	Summarize nightly build test failures	Manual	—	No

---

ap_debug_ci_failure

Single ReAct loop with one LLM (orchestrator). The agent investigates using pipeline MCP tools, reads the MR diff, and posts a review. Write operations require human approval.

graph TD
    A[agent] -->|tool calls| T[tools]
    T --> A
    A -->|no tool calls,\npost_mr_review not called| N[nudge]
    N --> A
    A -->|write tool call| HR[human_review\n— interrupt —]
    HR -->|approved| T
    HR -->|rejected + feedback| A
    HR -->|rejected| END
    A -->|no tool calls| END

---

ap_debug_ci_failure_multiagent

Multi-agent pipeline: triage identifies failures, parallel investigators diagnose each cluster, a synthesizer drafts the review, an optional code reviewer adds inline comments, and the review is posted after human approval.

graph TD
    TA[triage_agent] -->|tool calls| TT[triage_tools]
    TT --> TA
    TA -->|done| CJ[collect_job_info]
    CJ --> CL[cluster_jobs]

    CL -->|"Send() per cluster\n(parallel)"| IJ[investigate_job]
    CL -->|no clusters| SA

    IJ -->|optional| RF[review_findings\n— interrupt —]
    IJ --> SA[synthesize_agent]
    RF --> SA

    SA -->|tool calls| ST[synthesize_tools]
    ST --> SA
    SA -->|done| SRB[save_review_body]

    SRB -->|"MR + code review tools"| RC["review_code\n(Send)"]
    SRB -->|no code review| PPR
    SRB -->|no MR| END

    RC --> MCR[merge_code_review]
    MCR --> PPR[prepare_post_review]

    PPR --> HR[human_review\n— interrupt —]
    HR -->|approved| EPR[execute_post_review]
    HR -->|rejected + feedback| SA
    HR -->|rejected| END
    EPR --> END

    style HR fill:#f9f,stroke:#333
    style RF fill:#f9f,stroke:#333

---

ap_debug_stuck_productions

Single ReAct agent investigating a stuck production sample. Uses production MCP tools (sample details, MaxReset files, grid job logs). Write actions (adding known failures, creating GitLab issues) require human approval.

graph TD
    A[agent] -->|tool calls| T[tools]
    T --> A
    A -->|write tool call| HR[human_review\n— interrupt —]
    HR -->|approved| T
    HR -->|rejected + feedback| A
    HR -->|rejected| END
    A -->|no tool calls| END

---

mattermost_debug_user_request

Support bot triggered by admin in a Mattermost thread. Investigates using all available tools, drafts a response, then always pauses for admin verification before posting.

graph TD
    A[agent] -->|tool calls| T[tools]
    T --> A
    A -->|done| HV[human_verify\n— interrupt —]
    HV --> END

---

nightly_failure_summary

Simple ReAct agent that queries nightly build test results and produces a structured summary. Read-only — no interrupts.

graph TD
    A[agent] -->|tool calls| T[tools]
    T --> A
    A -->|done| END

---

Tools

GitLab tools (LbAgents.tools.gitlab)

Factory function make_gitlab_review_tools(gitlab_token, mr_iid) returns LangChain tools scoped to a specific merge request:

• get_mr_info() — MR metadata (title, author, branches, labels)
• get_mr_diff() — File-level diffs
• get_mr_file_content(file_path) — Full file from source branch
• post_mr_review(body, inline_comments) — Post review with inline comments
• get_mr_commits() — Commit list for the MR
• get_commit_diff(sha) — Diff for a specific commit
• get_target_branch_file(file_path) — File from the target branch

Nightly tools (LbAgents.tools.nightly)

Factory function make_nightly_tools(bucket_url) returns tools for querying nightly build results:

• get_nightly_status_summary(slot, build_id) — Overall build status
• list_nightly_platforms(slot, build_id) — Available platforms
• list_nightly_projects(slot, build_id, platform) — Projects for a platform
• get_failed_tests(slot, build_id, platform, project) — Failed test list
• get_test_output(slot, build_id, platform, project, test) — Test output
• get_test_measurement(slot, build_id, platform, project, test) — Performance data

Usage

from LbAgents import WORKFLOW_REGISTRY
from LbAgents.tools.gitlab import make_gitlab_review_tools

# All workflows are registered at import time
for name, spec in WORKFLOW_REGISTRY.items():
    print(f"{name}: {spec.description}")

# Build a graph with MCP tools + GitLab tools
from LbAgents.graphs import build_ap_debug_ci_failure_multiagent_graph

gitlab_tools = make_gitlab_review_tools(gitlab_token="...", mr_iid=42)
all_tools = mcp_tools + gitlab_tools

graph = build_ap_debug_ci_failure_multiagent_graph(
    orchestrator_llm=orchestrator,
    tools=all_tools,
    write_tool_names=["post_mr_review"],
    coder_llm=coder,  # optional, for code trace + review
)
compiled = graph.compile(checkpointer=my_checkpointer)

Adding a new workflow

1. Create a new module in src/LbAgents/graphs/ with a build_*_graph() factory function.
2. Add a state schema in src/LbAgents/schemas/state.py if needed.
3. Add a prompt template in src/LbAgents/prompts/templates.py.
4. Register the workflow in src/LbAgents/graphs/__init__.py via register_workflow().
5. Add tests in tests/test_graphs.py.

The workflow will automatically appear in LbAPI's GET /agent/workflows endpoint.

Development

pixi install
pixi run pytest        # Run tests
pixi run ruff check    # Lint
pixi run pre-commit

License

GPL-3.0 — see COPYING.