菜单
Surveys FeCIM domain knowledge by searching references/, citations/, docs/4-research/, and the local Cognee KG, then synthesizes a cited research note. Use when investigating a physics topic, evaluating a paper, or grounding a design decision in literature.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | fecim-researcher |
| description | Surveys FeCIM domain knowledge by searching references/, citations/, docs/4-research/, and the local Cognee KG, then synthesizes a cited research note. Use when investigating a physics topic, evaluating a paper, or grounding a design decision in literature. |
Survey FeCIM literature and project knowledge to ground a design decision or answer a physics question. See tools/fecim-skills/_shared/fecim-context.md for the canonical citation list.
Define the question. Write it in one sentence. If it has multiple sub-questions, split them and run the workflow per sub-question.
Search local sources in this order:
docs/4-research/ (audits, validation notes, error propagation)references/ (academic papers, simulation benchmarks)citations/ (project's citation registry, if present)experimental-data/ (HZO, HfO2, crossbar characterization)Use rg with focused patterns when available (e.g., rg -i "preisach" docs/4-research/ references/). If rg is missing, use grep -RIn or the file-search tool and state the fallback.
If Cognee is configured locally (.env has LLM_API_KEY, .cognee_system/ exists), query the KG:
python3 - <<'PY'
import os, asyncio
os.environ["COGNEE_SKIP_CONNECTION_TEST"] = "true"
os.environ["ENABLE_BACKEND_ACCESS_CONTROL"] = "false"
import cognee
async def main():
results = await cognee.search("YOUR QUERY HERE")
print(results)
asyncio.run(main())
PY
Otherwise skip it explicitly in the evidence log; do not install Cognee or start long bootstrap/download flows during active validation.
Cite findings using the canonical short forms from tools/fecim-skills/_shared/fecim-context.md. Never invent a citation; if no source supports a claim, label it "no source found, requires verification".
Output a structured note:
Question: <...>
Sources: <list with short citation forms>
Finding: <2-5 sentences>
Gaps: <what is unsupported, what should be measured>
Recommended next step: <action>
module1-hysteresis/pkg/ferroelectric/material.go and references/, cites Materlik 2015, returns parameters with units.Research output is observation — TDD: N/A. Any code change suggested by the research triggers the project's TDD hard-rule.
Audits Go code for goroutine-to-widget access without fyne.Do(...) wrapping, the project's most common GUI freeze cause. Use when reviewing a PR that adds goroutines, async I/O, or simulation tickers in any pkg/gui/ or shell package.
Runs build flows for the canonical Fyne shell and module commands on this Go 1.25 monorepo. Use when building, packaging, or debugging build failures in cmd/fecim-lattice-tools or module GUI commands.
Relentlessly interviews the user about a proposed FeCIM physics, simulation, or GUI change before any code is written — covering source citation, educational-vs-validated framing, TDD-RED test, thread-safety, and honesty-audit alignment. Use when starting a non-trivial change to physics, ISPP, crossbar, or GUI logic.
Runs the FeCIM test matrix (full, race, module-scoped, coverage, golden regen) and interprets physics regression failures using the 5 known bug patterns. Use when running tests, debugging test failures, or regenerating physics golden files.
Verifies and formats FeCIM physics/measurement claims against the project's published-source list and docs/4-research/honesty-audit.md. Use when adding a numeric claim, accuracy figure, or device-parameter assertion to code, docs, PR descriptions, or commit messages.
Enforces docs/4-research/honesty-audit.md policy by scanning PR diffs, READMEs, and presentation material for removed/unverified claims (87% MNIST, 30-states-as-fact, energy multipliers vs NAND/GPUs). Use before committing docs, PRs, or release notes that include accuracy or efficiency numbers.