メニュー
Lightweight search skill for the /search command. Instructs the agent to use flow.md files as a fast index, scope the search by tag (project or memory), and return a concise summary to the main agent.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Infinite improvement loop for any research artifact in any phase — a single managing agent makes one focused change per iteration, judges it against the previous best, keeps or reverts. Its memory is a per-loop wiki it reads before every move and writes after every move. The loop never stops until the human interrupts it. Inspired by Andrej Karpathy's autoresearch (MIT).
Phase guidance for /git — context-aware git shorthand rules, smart push/pull logic, commit message generation, branch management, and PR creation workflow.
This skill should be used whenever the user mentions BIDS, Brain Imaging Data Structure, BIDS conversion, BIDS validation, BIDS compliance, organizing neuroimaging data, dataset_description.json, participants.tsv, bids-validator, pybids, MNE-BIDS, or asks how to structure EEG/MEG/fMRI/iEEG/PET/DWI data for sharing or preprocessing. Also invoke when the user asks how to name scan files, what sidecar JSON fields are needed, how to set up derivatives/, or how to run fMRIPrep/MRIQC on their dataset. Invoke proactively during /data, /data-preprocess, and /data-analyze phases whenever the dataset structure is relevant to the task at hand.
Phase guidance for the neuroflow /data-analyze command. Loaded automatically when /data-analyze is invoked to orient agent behavior, relevant skills, and workflow hints for the data-analyze phase.
Phase guidance for the neuroflow /data-preprocess command. Loaded automatically when /data-preprocess is invoked to orient agent behavior, relevant skills, and workflow hints for the data-preprocess phase.
Phase guidance for the neuroflow /data command. Loaded automatically when /data is invoked to orient agent behavior, relevant skills, and workflow hints for the data intake phase.
| name | phase-search |
| description | Lightweight search skill for the /search command. Instructs the agent to use flow.md files as a fast index, scope the search by tag (project or memory), and return a concise summary to the main agent. |
Fast, minimal search of the project. Use this skill when running /search. Do not do deep exploration — read the index first, search only what is needed, and return a brief summary.
Every search invocation carries a scope tag. Recognise these two tags:
| Tag | Scope | What to search |
|---|---|---|
memory | Inside .neuroflow/ only | All .neuroflow/ files: project_config.md, flow.md, phase subfolders, sessions, reasoning |
project | Outside .neuroflow/ | Source files, scripts, notebooks, data files, config files in the repo root and subdirectories — everything except .neuroflow/ |
If no tag is provided, default to memory and inform the user.
memory searches.neuroflow/flow.md — this is the root index. Note which phase subfolders exist and their descriptions.flow.md — this lists the files and their one-line descriptions.project searches.neuroflow/flow.md first — it lists output_path values for each phase, pointing to where scripts, results, and manuscripts live.output_path entries to quickly locate the relevant directories.Return a compact summary, not a full report. Structure:
🔍 Search: "<query>" [scope: memory | project]
Found in:
• <file-path> — <one-line reason why it matches>
• <file-path> — <one-line reason why it matches>
Summary: <2–4 sentences synthesising what was found>
If nothing is found:
🔍 Search: "<query>" [scope: memory | project]
No matches found. Checked: <list of flow.md files consulted>
Suggestion: <one-line suggestion — try a different tag, different term, or check if the phase has been started>
.neuroflow/ or any project file.