atlas

Use when designing an ATLAS physics analysis end-to-end: choosing signal and control regions, background estimation strategy, selecting a ntupling framework (TopCPToolkit vs FastFrames vs coffea), selecting a statistical model (pyhf vs cabinetry vs TRExFitter vs HistFitter), verifying dataset availability via AMI/Rucio, designing a systematic uncertainty framework, planning an unblinding strategy, or producing a structured analysis specification document. Also use when asked "how should I approach this measurement/search" or "what framework should I use for my analysis".

Use when writing Python code to access, process, or plot ATLAS data: reading ROOT files with uproot, querying ATLAS datasets with ServiceX, manipulating event records with awkward-array, filling and styling histograms with hist/mplhep, computing physics quantities (invariant mass, deltaR, MET significance) with vector, or building an analysis pipeline from ntuples to plots. Handles ATLAS Open Data and full-collaboration datasets. Use atlas-analysis-architect first to produce a specification when starting a new analysis from scratch.

Use when discovering, locating, or inspecting ATLAS datasets: finding dataset containers by physics process or AMI tag, checking Rucio replica locations and availability, browsing ATLAS Open Data samples, listing branches in a ROOT file, or checking what variables are available in a DAOD. Also use when a user asks "what datasets exist for X", "where is my ttbar mc20 sample", or "what branches does this ROOT file have".

Use when answering questions about ATLAS software internals: Athena framework, CP algorithm setup, event data model (EDM), derivation formats (DAOD), ASG tools, CMake/ATLAS build system, CVMFS setup, analysis releases, or any topic covered by the ATLAS software documentation at atlas-software.docs.cern.ch. Also use when the user asks "how does X work in ATLAS software" or "where is the documentation for Y CP tool".

Use when designing or implementing an ATLAS statistical analysis: choosing between pyhf, cabinetry, HistFitter, TRExFitter, or RooUnfold, building a HistFactory workspace, defining nuisance parameters and systematics, running a profile likelihood fit, computing CLs exclusion limits, setting up an unfolding procedure, or debugging a fit that is not converging or has unexpected NP pulls.

Use when producing a structured ATLAS analysis specification document: formalizing an analysis design into a specification.md file, documenting signal/control/validation regions, background methods, systematic uncertainties, and statistical model choices, or when the atlas-analysis-architect agent needs to write its output to disk.

Use when a question involves ATLAS software concepts: Athena framework, event data model, CP algorithms, derivation formats (DAOD), CMake build system, CVMFS software releases, analysis releases, ASG tools, or any ATLAS-internal software infrastructure. This skill routes to the right resource or subagent; for deep lookup delegate to atlas-docs-expert.

Use when working with jagged or variable-length arrays in Python HEP analysis: building event records with ak.zip, filtering nested arrays, computing combinatorics (cartesian products or combinations), flattening ragged arrays, using argmin/argmax with keepdims, broadcasting per-event weights to per-object, or debugging OptionType/None-padding issues in awkward-array 2.x workflows.

Use when building an ATLAS statistical analysis with cabinetry: writing a cabinetry config file, building histogram templates from ROOT NTuples, constructing a pyhf workspace, running a profile likelihood fit, visualising pre/post-fit data-MC comparisons, producing pull plots and NP rankings, or computing CLs exclusion limits via cabinetry's high-level API.

Use when writing a columnar ATLAS analysis with coffea: defining a NanoEvents or custom processor, running over multiple files with dask-awkward or iterative executor, accumulating histograms with hist, applying scale factors and systematic weights, or migrating a for-loop event analysis to a coffea processor pattern.