ljx-chase/research-field-onboarding / field-onboarding
field-onboarding
VerifiedGuide a researcher step by step into an unfamiliar research field, or decode a paper, abstract, figure caption, or referee comment they cannot parse. Builds understanding in rungs (motivation, vocabulary, core framework, methods, frontier), anchored to what the user already knows, with a checkpoint before each advance. Use when the user says they are new to a field, asks what a research area or method is, says an explanation was too technical, asks to be walked through something step by step, asks for a reading path, or supplies dense research text. Trigger even when the user only names an unfamiliar field or pastes an abstract without asking to be taught. Do not trigger for a one-line summary, a short overview request, or a direct factual answer; do not use it for translation, editing, search, debugging, or specialist questions inside the user's own field. Also trigger in other languages, including Chinese such as 入门, 一步一步讲, 看不懂, 这篇论文讲什么, 帮我理解这个领域.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 18.0 KB |
| agents/openai.yaml | config | 139 B |
| references/anti-patterns.md | markdown | 3.6 KB |
| references/checkpoints.md | markdown | 2.1 KB |
| references/citations.md | markdown | 1.5 KB |
| references/decode-mode.md | markdown | 1.2 KB |
| references/evals.md | markdown | 16.0 KB |
| references/examples.md | markdown | 11.9 KB |
| references/explanation-styles.md | markdown | 3.3 KB |
| references/field-map.md | markdown | 4.4 KB |
| references/pacing.md | markdown | 2.1 KB |
| references/search-recipes.md | markdown | 3.2 KB |
| references/state-runtime.md | markdown | 4.0 KB |
| references/unsettled-fields.md | markdown | 2.4 KB |
| scripts/knowledge_state.py | script | 22.8 KB |
references/field-map.md
View on GitHub# Field map
Loaded after Step 0 of an onboarding request, before planning the path; at the
close; and when the user pastes a map back. Never build a map for a one-turn
answer, in Decode mode, after a declined ladder, or because a question looked
large.
## Map and path
The **map** is what the field contains and which parts depend on which, taken
from the structure Step 0 retrieved. It does not change during the session. The **path** is the three to six nodes this session walks, one turn
each, in dependency order, chosen by the target (`pacing.md`) and the Step 0
marks.
## The skeleton
Step 0's one search finds it. Take the first you can get: the standard graduate
textbook's table of contents, a recent review's section headings, or a
published course syllabus. Where a field has no textbook, as in much of the
humanities and many interdisciplinary fields, use a handbook's or companion
volume's contents, or a scholarly encyclopedia entry's sections.
Label it **verified** or **from memory, unverified**, as `citations.md` says.
Without search, also give one concrete action that checks it: the book and
edition whose contents page to compare, or the exact listing and query to run
(`search-recipes.md`). "Check a textbook" is not a check.
Never number a from-memory skeleton as a book's chapters, or give a chapter
count you did not retrieve: a fabricated contents page is worse than none,
because it looks checked.
A contents page gives order, not dependencies. Label every dependency you infer
**inferred**; only a source that states it makes it verified.
Keep the map at chapter level, six to twelve nodes. If the source has more,
merge neighbours and say so.
## Node states
- `new`: on the path, not yet reached;
- `taught`: covered in its node turn, not yet checked;
- `checked`: the user answered its checkpoint correctly;
- `shaky`: taught, but the answer was wrong or partial; re-teach it from a
different angle before its dependents;
- `skipped`: deliberately left out, with the reason in one clause.
Every node has exactly one. Nodes off the path start `skipped`, sharing a
clause when they share a reason; a node the user already has is skipped as
"already yours".
## Showing it
Do not ask about it in Step 0. Plan from it silently, and end Rung 1 with one
line saying the user can ask to see it.
When asked, show it once: the path one line per node with a one-clause gloss,
then the skipped nodes on a single line, a count and names only, grouped by
reason. Never reprint it; after each node turn, print only the nodes whose
state changed.
## Closing check and omissions
The session's last checkpoint tests structure, not recall: ask which node
depends on which, or why one node had to come before another. Do not ask for a
definition you already gave. The scoping rule in `checkpoints.md` still holds.
At the end, name every node you did not teach and what each one is needed for.
Do not present the path as though it were the whole field.
## Where the map lives
Within a session: the state helper when `state-runtime.md` allows it,
otherwise the conversation. Across sessions: only the closing artifact, which
the user saves and pastes back. Never imply that anything carries over on its
own.
When a map is pasted back, continue from its node states. Do not re-run the
Step 0 intake for nodes marked `taught` or `checked`. Keep the target and style
it records, re-teach `shaky` nodes first, and fold a one-line check on a
`taught` node the next turn needs into that turn's opening.
## Example
Single-cell RNA sequencing, for one paper on a new cell type. The closing
artifact, after the user stopped mid-path:
```text
Field map: single-cell RNA sequencing. [review] headings, verified, [DOI]
Target: one paper on a new cell type. Style: physical picture first
checked count matrix
shaky quality control: doublet removal still unclear
taught normalization
new dimensionality reduction: needed to judge what the UMAP shows
new clustering and annotation: needed for the paper's central claim
skipped differential expression (no conditions): needed to compare
conditions
skipped trajectory inference (no time order): needed for differentiation
Depends (inferred): counts -> QC -> normalization -> reduction ->
clustering -> DE; reduction -> trajectory
Paste this into a new session to continue.
```
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |