fmri
Coordinate end-to-end task-fMRI analyses across BIDS discovery, fmrireg models, fmrigds group inference, and neuromosaic reports. Use for multi-stage analysis plans or execution, not a single-package question.
- License 许可
- MIT
- 最近更新
- 2026-11-11
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- fmri
- license
- MIT
- metadata
- { "version": "0.1.0", "reviewed": "2026-09-27" }
- description
- Coordinate end-to-end task-fMRI analyses across BIDS discovery, fmrireg models, fmrigds group inference, and neuromosaic reports. Use for multi-stage analysis plans or execution, not a single-package question.
- compatibility
- Requires local filesystem access; R for package execution and Python 3.10+ for optional helpers.
02 资源文件 17
- 许可证LICENSE1.1 KB
- 参考文档references/artifact-contracts.md3.8 KB
- 参考文档references/capabilities.md2.7 KB
- 参考文档references/events-to-design.md9.9 KB
- 参考文档references/execution.md3.7 KB
- 参考文档references/interview.md4.3 KB
- 参考文档references/nifti-handoff.md9.9 KB
- 参考文档references/operating-contract.md5.1 KB
- 参考文档references/preferences.md3.5 KB
- 参考文档references/sources.md5.5 KB
- 参考文档references/trial-patterns-rsa.md4.9 KB
- 参考文档schemas/inventory.schema.json1.1 KB
- 参考文档schemas/plan.schema.json3.9 KB
- 参考文档schemas/preferences.schema.json1.4 KB
- 脚本scripts/probe_capabilities.R2.7 KB
- 脚本scripts/workbench.py20.8 KB
- 其他SHA256SUMS1.5 KB
03 SKILL.md 正文
fMRI workbench
Turn the user's scientific question into an inspectable, executable analysis.
Coordinate stages; do not replace their package APIs or load every reference.
For an isolated task, route directly to fmri-bids, fmrireg, fmrigds, or
neuromosaic and stop coordinating. Each is independently installable.
Start with evidence
Read the operating contract when creating or
changing an analysis. Reuse approved state and preferences before asking anything.
Discover which requested stages already have valid inputs. For datasets, use
fmri-bids to inspect local BIDS metadata first. Raw-only data can be inventoried
but these four packages do not implement a preprocessing pipeline. If the
independent fmriprep skill is installed and preprocessing is requested, use it
for that stage and record its downstream handoff manifest as
fingerprints.preprocessing_handoff (path, sha256) in plan.json; fit only
runs it lists as QC-passed. Otherwise record the
preprocessing handoff and missing capability; never fit raw BOLD or install a
preprocessor implicitly. Do not assume fMRIPrep completion means analysis readiness.
When the question is open, use events to candidate designs to infer plausible models from event tables and task documentation. Present a few supported candidates and their unresolved meanings before the interview. For trialwise patterns or representational comparisons, read the conditional trial-pattern/RSA guide and verify the required specialist capabilities before planning execution.
Present a compact observed / proposed / needs your decision summary. Resolve the scientific question, estimand, meaningful condition labels, cohort, and material ambiguities. Use the adaptive interview. Default to a standard review, no more than two rounds / six material questions when possible. Do not ask again for facts or choices already supplied. Interaction budget never licenses fabricated metadata or unapproved scientific decisions.
Plan, test, then execute
Save a plan envelope plus editable native R code. The execution protocol
links plan approval to code, inputs, environment and limits. Let package-native
objects carry implementation detail: fmrireg templates/jobs and fmrigds lazy plans.
Use artifact contracts for handoffs. Persist
what is known, proposed, confirmed, unresolved, or blocked; do not rely on chat
history for resumption. scripts/workbench.py --help exposes deterministic
state, audit, and preference helpers; it is not an analysis engine.
Read and invoke only the next relevant installed stage. Do not assume automatic cross-skill dependency installation. If a needed stage is absent, report the missing component; already available standalone work may still proceed. Keep parallelization subordinate to scientific dependencies. Run design checks and a small representative pilot before approved cohort execution. Review new scientific issues, not every reversible coding detail. Preserve partial failures.
Remember only with consent
Use preferences for conditional user/project profiles. A preference such as motion24 is not an instruction to silently omit unavailable columns. A B-spline baseline is not a B-spline HRF. Offer to save repeated choices, but write only after explicit consent. Protocol conflicts need an amendment.
Finish
Deliver the requested outputs, a readable methods/decision summary, QC/failure ledger, reproducibility information, and exact remaining limitations. Distinguish source review, executed numerical tests, and real-data validation. Do not present uncorrected display clusters as corrected findings or automatically exclude influential subjects. Keep report sharing and embedded voxel disclosure explicit. For environment differences, load capabilities.
04 同仓库更多 Skills
neuromosaic
Create neuromosaic fMRI map montages, atlas-annotated cluster reports, or interactive explorers from existing statistical outputs. Use without refitting analyses; distinguish display thresholds from statistical inference.
fmrireg
Specify, diagnose, fit, or export first-level task-fMRI GLMs with fmrireg, including HRFs, nuisance baselines, contrasts, and native batch jobs. Use independently of group analysis and reports; not for preprocessing.
fmrigds
Build and validate group-level fMRI analyses with fmrigds from existing effect/statistic maps or tables: covariates, uncertainty, repeated measures, corrections, and cohort examination. Does not require first-level refitting.
fmri
Coordinate end-to-end task-fMRI analyses across BIDS discovery, fmrireg models, fmrigds group inference, and neuromosaic reports. Use for multi-stage analysis plans or execution, not a single-package question.
fmri-bids
Inspect and audit BIDS fMRI datasets with bidser: inherited metadata, derivatives, events, confounds, and acquisition matching. Use for dataset discovery or analysis readiness; do not fit models or preprocess images.
neuromosaic
Create neuromosaic fMRI map montages, atlas-annotated cluster reports, or interactive explorers from existing statistical outputs. Use without refitting analyses; distinguish display thresholds from statistical inference.