Model ML Pipeline
This meta skill owns selection and ordering, not child methodology.
Do not load smoke-test-ml-pipeline as a sibling of evaluate.
Human-facing prose
Details: setup-workspace references/human_facing_prose.md.
Design notes, JOURNAL text, and # comments describe this
experiment — not the skills framework, the CLI, or the command that
produced an output. Questions, replies, and the close narrative
use the same data-science language — not skill ids, G-* names,
or the wrapper CLI. Trailing locator tokens stay index strings.
<!-- results-embed: … --> is a site marker. Authoring hints
stay in this skill. style is ruff only.
Design note shell
python -m skore_skills scaffold --journal --stem <NN_short> writes
journal/<stem>.md. Do not recreate it from memory. Fill Question,
Motivation, Method, and Risks as science (what to learn, why now,
what changes, what could invalidate the result). Status lifecycle:
planned → approved → running → done | abandoned. Those
four content sections freeze after approval; only Status, generated
notebook viewers, and the Method <!-- results-embed: pipeline -->
embed change afterwards. Abandoned: one-line reason on State;
Headline result n/a — abandoned: <reason>. There is no "Success
criteria" section. Keep ## Notebooks with Evaluation then Audit.
Entry routing — deterministic
Run python -m skore_skills status. If status.setup.pending
is non-empty and status.skills.setup-ml-project is true,
load setup-ml-project and stop. Do not start this skill.
When it returns, continue. Do not load it again on this turn.
If that skill is not installed, name the pending pieces in
one line and stop. Do not invent git init, scaffold, or
env init. If status.setup.env or status.setup.workspace
is declined, stop in one line. A declined git or
editable is not asked again and does not block modeling.
Framing is mandatory. If status.modeling_decisions is not
locked, load frame-ml-problem only if
status.skills.frame-ml-problem is true and stop. Do not invent
the table, offer model choices, write a design note, or declare
a pipeline. Missing skill → stop in one line. This includes an
approved-stem resume: do not build until the table is locked.
When the table is locked, run
python -m skore_skills frame show. Anything other than
proceed loads frame-ml-problem and stops. A proceed whose
translation is null has no splitter translation: say so and
stop. Do not write model code. Do not present a choice list:
no dummy predictor, standard baseline, EDA-driven proposal,
Backlog row, or discussion.
Resume beats menu. If the user names an experiment stem, or
status.policy.loop.stem / last_history_stem identifies a
current design, run
python -m skore_skills design consent --stem <stem>. Treat
JSON action as authoritative. proceed → resume that stem
directly; do not ask how to start again. ask → Design
approval below (render JSON context inline per § Gate
context); do not write code. stop → missing
note: name scaffold --journal --stem (or abandoned: explain
and do not implement). Do not infer approval from "build it".
Missing shell: no site build, no fill from memory.
Otherwise run python -m skore_skills model choices. Treat its
JSON as authoritative.
action stop / modeling_decisions_unlocked — load
frame-ml-problem and stop. Do not offer a model menu.
- If
choices includes baseline and the user did not ask to
discuss, propose a pipeline from the EDA, or pick a backlog
item, do not AskUserQuestion. Take the locked-baseline
contract below. On this turn, run that contract in order —
python -m skore_skills status, python -m skore_skills model choices, python -m skore_skills scaffold --journal --stem <NN_short>, then fill the note — before any
design-note prose or question. The only gate after the note
is the three lines in the locked-baseline contract, not
Yes / No.
- An explicit discuss, EDA, or backlog request takes that
branch even when
baseline is present. Do not show the menu
first.
- Otherwise present exactly
choices[], in returned order, in
one single-choice AskUserQuestion. Drop baseline from
that question: it is not a menu item.
eda_proposal → Propose a pipeline from the EDA
backlog → Pick from the Backlog
discuss → Discuss the next step
Carry each choice's JSON reason as its description so the user
reads why it is on offer. Choosing backlog asks which row, not
whether to build it.
Do not add a disabled choice, infer availability yourself, or
reorder the list. In particular: no locked-baseline choice when
model_stems is non-empty; no EDA proposal unless
data_analysis is present; no Backlog option when backlog is
empty. Discussion is present only after the lock.
Gate context
Every approval question carries its own context. Before asking,
state in 2–4 lines what the answer authorizes, the facts it rests
on — echoed inline — and what each option does, in data-science
terms (Approve / Modify / Stop; dummy vs baseline). On the
locked-baseline branch, those lines name the baseline and its
note only, and the options are the three lines in that contract.
Do not name
skill ids or G-* tokens in the question. A file link is an
addition, never the context: "read journal/<stem>.md and
approve" is not an approval request.
On design consent ask, the JSON context holds those facts:
question, source, files_touched, change, and up to two
risks. Quote them in the message (question and planned change
first, then the risks the user may push back on). After the note
is populated, and before Approve / Modify / Stop, run
python -m skore_skills site build --if-stale when site policy
is true. In that same question, say the website has been
updated and paste the file:// line from that output
verbatim. Do not name the
design-note markdown or html/<stem>.html. An empty
field means the note does not state that fact: say
the note is not ready for approval and offer to populate it;
never fill it from memory. Do not site build an empty shell.
Choice contracts
An EDA proposal or a discussion first produces a user-confirmed
proposal and an approved design note. The proposal yes agrees
that idea. A locked baseline and a backlog row do not: the frame
table or the picked row is the decision. The note is approved
only by Design approval below. No branch writes model code before
that gate is proceed. Use the next available numeric stem; never
overwrite an existing note.
Locked baseline (baseline). The note names the one
comparison model the journal already locked. A dummy token is a
DummyClassifier or DummyRegressor inside the normal skrub
DataOps declaration: it proves loading, fit/predict, and pytest
smoke, and it is not expected to add predictive value. Any
other token (logistic, seasonal_naive, group_mean,
production) is that one comparison model. Do not upgrade it
to another estimator. Do not AskUserQuestion before the note.
The message lists these three commands, in order, before any
design-note text. A workspace summary is not a substitute:
python -m skore_skills status
python -m skore_skills model choices
python -m skore_skills scaffold --journal --stem <NN_short>
Then fill Question, Motivation, Method, and Risks from the
locked baseline, its note, and the comparison metric. Before
the question, two to four lines name that baseline and its
note. They do not say what Approve allows. The question is
only these three lines:
- Approve — accept this note
- Modify — change this note
- Stop — leave this note unchanged
Nothing follows those lines. Do not mention discuss, an EDA
proposal, or the backlog unless the user already asked for one
of those. Keep the normal post-smoke Evaluate (Recommended) /
Modify / Stop gate for a later turn.
EDA proposal (eda_proposal). Read
data_analysis/data_analysis.md and the project goal. Cite the
EDA findings that motivate one pipeline proposal. Do not invent
findings or present multiple silent alternatives. Restate the
proposal, then one single-choice AskUserQuestion: Yes /
No. Do not also ask for a typed yes. "Maybe" is not Yes.
No or Stop writes nothing. A later Yes (the tool answer, or an
explicit yes on a later turn) is what authorizes the write.
On Yes, write the note, then Design approval, before build.
Backlog (backlog). Load manage-ml-backlog only if
status.skills.manage-ml-backlog is true. Else one-line skip;
do not invent a Backlog. Pass the backlog rows returned by
the CLI. That skill asks for one B<N> when the user has not
named one, and asks only for shaping facts the row does not
state. It does not ask Yes / No. When it returns the proposal,
run scaffold --journal --stem <NN_short> and fill the note
from that Item, Source, and the facts just given, then Design
approval, before build. Do not add an estimator or a procedure
neither the row nor those facts stated. Do not add a new
Backlog idea in this branch.
Discussion (discuss). Have an open conversation about what
to learn, why now, and what changes. Restate the agreed idea,
then one single-choice AskUserQuestion: Yes / No.
Do not also ask for a typed yes. "Maybe" is not Yes. No or Stop
writes nothing. A later Yes (the tool answer, or an explicit
yes on a later turn) is what authorizes the write. Only after
that confirmation create/populate the design note, then Design
approval. If no idea is agreed, return to the entry choices.
Before execution
This dispatcher labels the next kind of work but does not
duplicate a child's detailed preview.
- For
discuss, proposal shaping, or literature-backed design
work, emit 1–3 natural sentences saying this is LLM
discussion/research over recorded project facts, with no
model fit, smoke test, or CV. Name any scratch research note or
design note that may be written and stop at confirmation.
- For an approved implementation, say which child comes next and
the broad sequence: local pipeline preparation → small
real-data smoke fit/predict → optional full-dataset evaluation
→ review. Then let
build-ml-pipeline,
smoke-test-ml-pipeline, evaluate-ml-pipeline, and
review-ml-experiment each own the single detailed Before
execution preview at its actual compute boundary. The dispatcher
preview is those four phase names only, one short line each. Do
not mention the Method viewer, a site rebuild, a row-count
assertion, or skore.evaluate. Do not list
their commands (status, frame show, design consent,
smoke run) or a DataOps declaration in this preview. Do not
name loop notebooks or notebook convert. Notebooks stay
off until a report exists. With no shell, that four-phase
paragraph is the whole answer: do not name the post-smoke
Evaluate / Modify / Stop question, an unfitted site build,
or pipeline.svg.
Do not invent minute estimates at dispatcher level. Name a known
duration only when explicit measured evidence is available;
otherwise leave cost details to the child that knows data scope,
folds, and selected views. Pending proposal/design approval may
preview the sequence but never starts local work.
If the design-note shell is missing, this turn only names
python -m skore_skills scaffold --journal --stem <NN_short>
and stops, except on the locked-baseline branch and on a backlog
proposal that skill already returned. Those two run the command
and populate Question / Motivation / Method / Risks in this same
turn, then Design approval. Do not fill those sections from
memory when the command did not run this turn.
Design approval
One gate approves a populated note. Run
python -m skore_skills design consent --stem <stem>.
proceed → already approved; do not ask again. ask → render
the JSON context inline (§ Gate context), including the site
preview when policy.site, then AskUserQuestion (single
choice), in order: Approve / Modify / Stop. Do not
also ask in chat whether the note looks right.
- Approve → run
python -m skore_skills design approve --stem <stem>.
Do not type the date. Re-run design consent; code starts
only on proceed.
- Modify → leave
State planned, edit the note, and ask
this gate again.
- Stop → do not implement.
Approved-design implement loop
- Load
build-ml-pipeline only if status.skills.build-ml-pipeline
is true; else one-line skip and stop — do not declare the
pipeline from this meta. That skill loads smoke-test-ml-pipeline
after the experiment file exists and runs
python -m skore_skills smoke run --stem <stem>. JSON stop
stays in build (modify the pipeline, re-run smoke run).
Do not tell the user to edit the smoke test's expected row
count so it matches the sample.
proceed: build reports the
design, then
python -m skore_skills evaluate consent --stem <stem>
(Evaluate / Modify / Stop on ask, with that JSON context
rendered inline per § Gate context). Build also writes the
unfitted Method viewer
(scratch/results/<stem>/pipeline.svg and the DataOp report
scratch/results/<stem>/pipeline/) and, when
policy.site is true and export-ml-site is installed, runs
site build --if-stale after that snapshot and before Evaluate so Method
shows that report. Missing or skipped EDA does not defer it.
The post-loop rebuild refreshes the same unevaluated report;
it does not replace this one. Do not fill
experiments/<stem>.py at this unfitted snapshot if it
already contains skore.evaluate. That ban ends when
loop notebooks returns convert.
- Only if the user chose Evaluate and
smoke run is proceed: load
evaluate-ml-pipeline only if status.skills.evaluate-ml-pipeline
is true. It reuses the DataOp cv and writes
skore.evaluate in the experiment script. Re-run status
first, then python -m skore_skills evaluate consent --stem <stem>. Consent JSON is authoritative, not the user's wording
alone. Missing evaluate-ml-pipeline → one-line skip. Do not
invent that skill's steps.
- After a successful dispatched evaluate (locator returned): run
python -m skore_skills review consent --stem <stem>. Treat
JSON action as authoritative.
stop — no scratch/results/<stem>/report.html. Do not
review. Name that file. This is the evaluation snapshot, not
the site launcher report.html. Do not record-outcome.
audit or proceed — load review-ml-experiment only if
status.skills.review-ml-experiment is true. It returns the
digest, G-AUDIT-FINDING, locator, and idea paths. Do not
load audit-ml-pipeline from this dispatcher.
Then § Close, in that order.
- Missing skill → one-line skip and run § Close with
n/a — audit not run.
Do not duplicate child-skill methodology. Before new library
symbols are written, children use
python -m skore_skills api get <dotted>.... Several symbols
go in one call.
Stop conditions
- Pending setup (
status.setup.pending non-empty): load
setup-ml-project and do not start build. A declined git
is not asked again. A declined env or workspace stops.
- Generic model landing: do not hand-author the menu; run
python -m skore_skills model choices.
- If
journal/NN_<short>.md is missing, this turn only names
python -m skore_skills scaffold --journal --stem <NN_short> and stops, except the locked baseline and a
returned backlog proposal. Those run that command and populate
Question / Motivation / Method / Risks in the same turn, then
Design approval. Do not fill the template from memory when the
command did not run. Do not site build before the shell exists.
- Require
design consent proceed before code. Do not treat
"the user asked to build", or a chat yes on the drafted note,
as approval.
- Do not open an approval gate whose only context is a file path;
render the
context facts inline (§ Gate context).
- Preserve identical stems across design, experiment, smoke, audit.
- Do not replace skrub DataOps with bare sklearn Pipeline.
- Do not persist a result as done while
smoke run is stop.
- Do not load evaluate (or write
skore.evaluate) before the
post-smoke HITL answer is Evaluate (evaluate consent ask),
except proceed for a stem that already has a persisted report,
or while smoke run is stop.
- Do not load
smoke-test-ml-pipeline from this dispatcher.
- Do not duplicate child-skill methodology in this dispatcher.
- If
status.data_analysis is missing, continue with facts the user
stated; do not invent an exploratory data analysis report.
- Literature-backed feature-engineering or learner-family
questions: load
build-ml-pipeline only if
status.skills.build-ml-pipeline is true (it loads
research-ml-practice only if status.skills.research-ml-practice
is true). Else one-line skip. Do not distill research
here; do not invent papers from memory.
Close
Reached only after review returns, a missing review skill, or
red smoke. The four-phase preview does not enter here, and
neither does review consent stop. Run the steps in
this order. Do not reorder them. loop notebooks continues only
on skip. record is the loop artifacts action, not this
gate. Red smoke skips steps 3 and 4: do not mark the
experiment done. This dispatcher owns the user-facing close.
Children return locator / digest / finding and do not preview it.
python -m skore_skills loop notebooks --stem <stem>. Treat
JSON action as authoritative. Do not record-outcome,
site build, or git end-turn while action is convert.
- On
convert, run
python -m skore_skills notebook fill <source> for every
experiments/ or audit/ entry, with --html when html
is true. Do not notebook convert those files. Fill does not
put. Re-run the matching materialize.py first only when
that human call changed since the run that wrote the current
artifacts. Filling only audit/<stem>.py does not finish the
close. Re-run step 1 until action is skip. Missing
jupytext / nbformat / nbconvert → one-line skip naming
add-python-package; do not fail the turn. skip /
not_evaluated means the report is missing: do not fill an
unfitted script. When action is convert, the
unfitted-snapshot ban does not apply, including when
experiments/<stem>.py already contains skore.evaluate.
- record-outcome, only once step 1 is
skip. Load
manage-ml-backlog only if
status.skills.manage-ml-backlog is true, in record-outcome
mode, handing it the locator evaluate already returned,
plus the optional headline and G-AUDIT-FINDING
(n/a — audit not run when skipped). Else one-line skip; do
not write History from this meta. It writes the
journal/JOURNAL.md History row and design-note Status block
plus ## Results, then returns. It does not triage idea
files. Never mark done while smoke run is stop. Missing
headline becomes n/a, never an invented metric.
site build only when policy.site is true and
export-ml-site is installed:
python -m skore_skills site build --if-stale, so the Method DataOp
report and Results replace the construct-time snapshot. Skip
in one line otherwise. Site build embeds
audit/<stem>.nb.html under ## Notebooks; do not add
<!-- results-embed: audit -->. If site build errors with
mkdocs-material is required, load add-python-package for
mkdocs-material (agent) and build once more. Do not
pixi add / uv add. Name a build error; do not fail the
turn.
- User-facing close. 2–6 sentences of the result, grounded in
the audit digest when present (Checks + Metrics), else the
user's headline /
report.txt. Do not invent a metric. When
policy.site is true and the build ran, say the website has
been updated and paste the file:// line from that output
verbatim. Do not name
html/<stem>.html or
journal/<stem>.md. When the site was not built, do not
claim it was updated and do not name the markdown. Then
G-REPORT-LOCATOR (the locator evaluate passed up, or
n/a — backend did not expose a locator) first among
tokens, then G-AUDIT-FINDING. Index strings, not the narrative.
python -m skore_skills git end-turn --stage implement. If
JSON action is invoke, load persist-ml-git only if
status.skills.persist-ml-git is true and stop; that skill
returns to triage. If persist is missing, name the pending
staged paths and stop. Otherwise load triage-ml-task only
if status.skills.triage-ml-task is true; else stop. Do not
run git commit in this skill.
1---2name: model-ml-pipeline3description: >4 Require locked problem framing, write the first baseline design note for5 approval, then coordinate later choices, build, smoke testing, evaluation,6 and audit. On a first landing, write the locked baseline design note and ask7 for approval. Later landings offer only the choices the workspace justifies8 (an EDA proposal, Backlog, discussion). A missing modeling lock loads9 `frame-ml-problem` and stops. After a design is approved, coordinate build10 (pytest smoke is a build sub-step), the user's Evaluate (Recommended) /11 Modify / Stop gate, evaluation, and audit. Not for a single action already12 owned by evaluate, audit, or smoke debugging.13metadata:14 source: https://skillmd.com/skills/probabl-ai/model-ml-pipeline15 modelTier: big16---1718# Model ML Pipeline1920This meta skill owns selection and ordering, not child methodology.21Do not load `smoke-test-ml-pipeline` as a sibling of evaluate.2223## Human-facing prose2425Details: `setup-workspace` `references/human_facing_prose.md`.26Design notes, JOURNAL text, and `#` comments describe **this**27experiment — not the skills framework, the CLI, or the command that28produced an output. Questions, replies, and the close narrative29use the same data-science language — not skill ids, `G-*` names,30or the wrapper CLI. Trailing locator tokens stay index strings.31`<!-- results-embed: … -->` is a site marker. Authoring hints32stay in this skill. `style` is ruff only.3334## Design note shell3536`python -m skore_skills scaffold --journal --stem <NN_short>` writes37`journal/<stem>.md`. Do not recreate it from memory. Fill Question,38Motivation, Method, and Risks as science (what to learn, why now,39what changes, what could invalidate the result). Status lifecycle:40`planned` → `approved` → `running` → `done` | `abandoned`. Those41four content sections freeze after approval; only Status, generated42notebook viewers, and the Method `<!-- results-embed: pipeline -->`43embed change afterwards. Abandoned: one-line reason on State;44Headline result `n/a — abandoned: <reason>`. There is no "Success45criteria" section. Keep `## Notebooks` with Evaluation then Audit.4647## Entry routing — deterministic48491. Run `python -m skore_skills status`. If `status.setup.pending`50 is non-empty and `status.skills.setup-ml-project` is true,51 load `setup-ml-project` and stop. Do not start this skill.52 When it returns, continue. Do not load it again on this turn.53 If that skill is not installed, name the pending pieces in54 one line and stop. Do not invent `git init`, scaffold, or55 `env init`. If `status.setup.env` or `status.setup.workspace`56 is `declined`, stop in one line. A declined `git` or57 `editable` is not asked again and does not block modeling.582. **Framing is mandatory.** If `status.modeling_decisions` is not59 `locked`, load `frame-ml-problem` only if60 `status.skills.frame-ml-problem` is true and stop. Do not invent61 the table, offer model choices, write a design note, or declare62 a pipeline. Missing skill → stop in one line. This includes an63 approved-stem resume: do not build until the table is locked.64 When the table is `locked`, run65 `python -m skore_skills frame show`. Anything other than66 `proceed` loads `frame-ml-problem` and stops. A `proceed` whose67 `translation` is null has no splitter translation: say so and68 stop. Do not write model code. Do not present a choice list:69 no dummy predictor, standard baseline, EDA-driven proposal,70 Backlog row, or discussion.713. **Resume beats menu.** If the user names an experiment stem, or72 `status.policy.loop.stem` / `last_history_stem` identifies a73 current design, run74 `python -m skore_skills design consent --stem <stem>`. Treat75 JSON `action` as authoritative. `proceed` → resume that stem76 directly; do not ask how to start again. `ask` → Design77 approval below (render JSON `context` inline per § Gate78 context); do not write code. `stop` → missing79 note: name `scaffold --journal --stem` (or abandoned: explain80 and do not implement). Do not infer approval from "build it".81 Missing shell: no `site build`, no fill from memory.824. Otherwise run `python -m skore_skills model choices`. Treat its83 JSON as authoritative.84 - `action` `stop` / `modeling_decisions_unlocked` — load85 `frame-ml-problem` and stop. Do not offer a model menu.86 - If `choices` includes `baseline` and the user did not ask to87 discuss, propose a pipeline from the EDA, or pick a backlog88 item, do not AskUserQuestion. Take the locked-baseline89 contract below. On this turn, run that contract in order —90 `python -m skore_skills status`, `python -m skore_skills91 model choices`, `python -m skore_skills scaffold --journal92 --stem <NN_short>`, then fill the note — before any93 design-note prose or question. The only gate after the note94 is the three lines in the locked-baseline contract, not95 Yes / No.96 - An explicit discuss, EDA, or backlog request takes that97 branch even when `baseline` is present. Do not show the menu98 first.99 - Otherwise present exactly `choices[]`, in returned order, in100 one single-choice **AskUserQuestion**. Drop `baseline` from101 that question: it is not a menu item.102 - `eda_proposal` → **Propose a pipeline from the EDA**103 - `backlog` → **Pick from the Backlog**104 - `discuss` → **Discuss the next step**105106 Carry each choice's JSON `reason` as its description so the user107 reads why it is on offer. Choosing backlog asks which row, not108 whether to build it.109110Do not add a disabled choice, infer availability yourself, or111reorder the list. In particular: no locked-baseline choice when112`model_stems` is non-empty; no EDA proposal unless113`data_analysis` is `present`; no Backlog option when `backlog` is114empty. Discussion is present only after the lock.115116## Gate context117118Every approval question carries its own context. Before asking,119state in 2–4 lines what the answer authorizes, the facts it rests120on — echoed inline — and what each option does, in data-science121terms (Approve / Modify / Stop; dummy vs baseline). On the122locked-baseline branch, those lines name the baseline and its123note only, and the options are the three lines in that contract.124Do not name125skill ids or `G-*` tokens in the question. A file link is an126addition, never the context: "read `journal/<stem>.md` and127approve" is not an approval request.128129On `design consent` `ask`, the JSON `context` holds those facts:130`question`, `source`, `files_touched`, `change`, and up to two131`risks`. Quote them in the message (question and planned change132first, then the risks the user may push back on). After the note133is populated, and before **Approve** / **Modify** / **Stop**, run134`python -m skore_skills site build --if-stale` when site policy135is true. In that same question, say the website has been136updated and paste the `file://` line from that output137verbatim. Do not name the138design-note markdown or `html/<stem>.html`. An empty139field means the note does not state that fact: say140the note is not ready for approval and offer to populate it;141never fill it from memory. Do not `site build` an empty shell.142143## Choice contracts144145An EDA proposal or a discussion first produces a user-confirmed146proposal and an approved design note. The proposal yes agrees147that idea. A locked baseline and a backlog row do not: the frame148table or the picked row is the decision. The note is approved149only by Design approval below. No branch writes model code before150that gate is `proceed`. Use the next available numeric stem; never151overwrite an existing note.152153- **Locked baseline (`baseline`).** The note names the one154 comparison model the journal already locked. A `dummy` token is a155 `DummyClassifier` or `DummyRegressor` inside the normal skrub156 DataOps declaration: it proves loading, fit/predict, and pytest157 smoke, and it is not expected to add predictive value. Any158 other token (`logistic`, `seasonal_naive`, `group_mean`,159 `production`) is that one comparison model. Do not upgrade it160 to another estimator. Do not AskUserQuestion before the note.161 The message lists these three commands, in order, before any162 design-note text. A workspace summary is not a substitute:163164 ```165 python -m skore_skills status166 python -m skore_skills model choices167 python -m skore_skills scaffold --journal --stem <NN_short>168 ```169170 Then fill Question, Motivation, Method, and Risks from the171 locked baseline, its note, and the comparison metric. Before172 the question, two to four lines name that baseline and its173 note. They do not say what Approve allows. The question is174 only these three lines:175176 - **Approve** — accept this note177 - **Modify** — change this note178 - **Stop** — leave this note unchanged179180 Nothing follows those lines. Do not mention discuss, an EDA181 proposal, or the backlog unless the user already asked for one182 of those. Keep the normal post-smoke Evaluate (Recommended) /183 Modify / Stop gate for a later turn.184- **EDA proposal (`eda_proposal`).** Read185 `data_analysis/data_analysis.md` and the project goal. Cite the186 EDA findings that motivate one pipeline proposal. Do not invent187 findings or present multiple silent alternatives. Restate the188 proposal, then one single-choice **AskUserQuestion**: **Yes** /189 **No**. Do not also ask for a typed yes. "Maybe" is not Yes.190 No or Stop writes nothing. A later Yes (the tool answer, or an191 explicit yes on a later turn) is what authorizes the write.192 On Yes, write the note, then Design approval, before build.193- **Backlog (`backlog`).** Load `manage-ml-backlog` only if194 `status.skills.manage-ml-backlog` is true. Else one-line skip;195 do not invent a Backlog. Pass the `backlog` rows returned by196 the CLI. That skill asks for one `B<N>` when the user has not197 named one, and asks only for shaping facts the row does not198 state. It does not ask Yes / No. When it returns the proposal,199 run `scaffold --journal --stem <NN_short>` and fill the note200 from that Item, Source, and the facts just given, then Design201 approval, before build. Do not add an estimator or a procedure202 neither the row nor those facts stated. Do not add a new203 Backlog idea in this branch.204- **Discussion (`discuss`).** Have an open conversation about what205 to learn, why now, and what changes. Restate the agreed idea,206 then one single-choice **AskUserQuestion**: **Yes** / **No**.207 Do not also ask for a typed yes. "Maybe" is not Yes. No or Stop208 writes nothing. A later Yes (the tool answer, or an explicit209 yes on a later turn) is what authorizes the write. Only after210 that confirmation create/populate the design note, then Design211 approval. If no idea is agreed, return to the entry choices.212213## Before execution214215This dispatcher labels the next kind of work but does not216duplicate a child's detailed preview.217218- For `discuss`, proposal shaping, or literature-backed design219 work, emit 1–3 natural sentences saying this is **LLM220 discussion/research** over recorded project facts, with no221 model fit, smoke test, or CV. Name any scratch research note or222 design note that may be written and stop at confirmation.223- For an approved implementation, say which child comes next and224 the broad sequence: local pipeline preparation → small225 real-data smoke fit/predict → optional full-dataset evaluation226 → review. Then let `build-ml-pipeline`,227 `smoke-test-ml-pipeline`, `evaluate-ml-pipeline`, and228 `review-ml-experiment` each own the single detailed Before229 execution preview at its actual compute boundary. The dispatcher230 preview is those four phase names only, one short line each. Do231 not mention the Method viewer, a site rebuild, a row-count232 assertion, or `skore.evaluate`. Do not list233 their commands (`status`, `frame show`, `design consent`,234 `smoke run`) or a DataOps declaration in this preview. Do not235 name `loop notebooks` or `notebook convert`. Notebooks stay236 off until a report exists. With no shell, that four-phase237 paragraph is the whole answer: do not name the post-smoke238 Evaluate / Modify / Stop question, an unfitted `site build`,239 or `pipeline.svg`.240241Do not invent minute estimates at dispatcher level. Name a known242duration only when explicit measured evidence is available;243otherwise leave cost details to the child that knows data scope,244folds, and selected views. Pending proposal/design approval may245preview the sequence but never starts local work.246247If the design-note shell is missing, this turn only names248`python -m skore_skills scaffold --journal --stem <NN_short>`249and stops, except on the locked-baseline branch and on a backlog250proposal that skill already returned. Those two run the command251and populate Question / Motivation / Method / Risks in this same252turn, then Design approval. Do not fill those sections from253memory when the command did not run this turn.254255## Design approval256257One gate approves a populated note. Run258`python -m skore_skills design consent --stem <stem>`.259`proceed` → already approved; do not ask again. `ask` → render260the JSON `context` inline (§ Gate context), including the site261preview when `policy.site`, then **AskUserQuestion** (single262choice), in order: **Approve** / **Modify** / **Stop**. Do not263also ask in chat whether the note looks right.264- **Approve** → run265 `python -m skore_skills design approve --stem <stem>`.266 Do not type the date. Re-run `design consent`; code starts267 only on `proceed`.268- **Modify** → leave `State` `planned`, edit the note, and ask269 this gate again.270- **Stop** → do not implement.271272## Approved-design implement loop2732741. Load `build-ml-pipeline` only if `status.skills.build-ml-pipeline`275 is true; else one-line skip and stop — do not declare the276 pipeline from this meta. That skill loads `smoke-test-ml-pipeline`277 after the experiment file exists and runs278 `python -m skore_skills smoke run --stem <stem>`. JSON `stop`279 stays in build (modify the pipeline, re-run `smoke run`).280 Do not tell the user to edit the smoke test's expected row281 count so it matches the sample.282 `proceed`: build reports the283 design, then284 `python -m skore_skills evaluate consent --stem <stem>`285 (Evaluate / Modify / Stop on `ask`, with that JSON `context`286 rendered inline per § Gate context). Build also writes the287 unfitted Method viewer288 (`scratch/results/<stem>/pipeline.svg` and the DataOp report289 `scratch/results/<stem>/pipeline/`) and, when290 `policy.site` is true and `export-ml-site` is installed, runs291 `site build --if-stale` after that snapshot and before Evaluate so Method292 shows that report. Missing or skipped EDA does not defer it.293 The post-loop rebuild refreshes the same unevaluated report;294 it does not replace this one. Do not fill295 `experiments/<stem>.py` at this unfitted snapshot if it296 already contains `skore.evaluate`. That ban ends when297 `loop notebooks` returns `convert`.2982. Only if the user chose **Evaluate** and `smoke run` is `proceed`: load299 `evaluate-ml-pipeline` only if `status.skills.evaluate-ml-pipeline`300 is true. It reuses the DataOp `cv` and writes301 `skore.evaluate` in the experiment script. Re-run `status`302 first, then `python -m skore_skills evaluate consent --stem303 <stem>`. Consent JSON is authoritative, not the user's wording304 alone. Missing `evaluate-ml-pipeline` → one-line skip. Do not305 invent that skill's steps.3063. After a successful dispatched evaluate (locator returned): run307 `python -m skore_skills review consent --stem <stem>`. Treat308 JSON `action` as authoritative.309 - `stop` — no `scratch/results/<stem>/report.html`. Do not310 review. Name that file. This is the evaluation snapshot, not311 the site launcher `report.html`. Do not record-outcome.312 - `audit` or `proceed` — load `review-ml-experiment` only if313 `status.skills.review-ml-experiment` is true. It returns the314 digest, G-AUDIT-FINDING, locator, and idea paths. Do not315 load `audit-ml-pipeline` from this dispatcher.316 Then § Close, in that order.317 - Missing skill → one-line skip and run § Close with318 `n/a — audit not run`.319320Do not duplicate child-skill methodology. Before new library321symbols are written, children use322`python -m skore_skills api get <dotted>...`. Several symbols323go in one call.324325## Stop conditions326327- Pending setup (`status.setup.pending` non-empty): load328 `setup-ml-project` and do not start build. A declined `git`329 is not asked again. A declined env or workspace stops.330- Generic model landing: do not hand-author the menu; run331 `python -m skore_skills model choices`.332- If `journal/NN_<short>.md` is missing, this turn only names333 `python -m skore_skills scaffold --journal --stem334 <NN_short>` and stops, except the locked baseline and a335 returned backlog proposal. Those run that command and populate336 Question / Motivation / Method / Risks in the same turn, then337 Design approval. Do not fill the template from memory when the338 command did not run. Do not `site build` before the shell exists.339- Require `design consent` `proceed` before code. Do not treat340 "the user asked to build", or a chat yes on the drafted note,341 as approval.342- Do not open an approval gate whose only context is a file path;343 render the `context` facts inline (§ Gate context).344- Preserve identical stems across design, experiment, smoke, audit.345- Do not replace skrub DataOps with bare sklearn Pipeline.346- Do not persist a result as done while `smoke run` is `stop`.347- Do not load evaluate (or write `skore.evaluate`) before the348 post-smoke HITL answer is Evaluate (`evaluate consent` `ask`),349 except `proceed` for a stem that already has a persisted report,350 or while `smoke run` is `stop`.351- Do not load `smoke-test-ml-pipeline` from this dispatcher.352- Do not duplicate child-skill methodology in this dispatcher.353- If `status.data_analysis` is `missing`, continue with facts the user354 stated; do not invent an exploratory data analysis report.355- Literature-backed feature-engineering or learner-family356 questions: load `build-ml-pipeline` only if357 `status.skills.build-ml-pipeline` is true (it loads358 `research-ml-practice` only if `status.skills.research-ml-practice`359 is true). Else one-line skip. Do not distill research360 here; do not invent papers from memory.361362## Close363364Reached only after review returns, a missing review skill, or365red smoke. The four-phase preview does not enter here, and366neither does `review consent` `stop`. Run the steps in367this order. Do not reorder them. `loop notebooks` continues only368on `skip`. `record` is the `loop artifacts` action, not this369gate. Red smoke skips steps 3 and 4: do not mark the370experiment done. This dispatcher owns the user-facing close.371Children return locator / digest / finding and do not preview it.3723731. `python -m skore_skills loop notebooks --stem <stem>`. Treat374 JSON `action` as authoritative. Do not record-outcome,375 `site build`, or `git end-turn` while `action` is `convert`.3762. On `convert`, run377 `python -m skore_skills notebook fill <source>` for every378 `experiments/` or `audit/` entry, with `--html` when `html`379 is true. Do not `notebook convert` those files. Fill does not380 `put`. Re-run the matching `materialize.py` first only when381 that human call changed since the run that wrote the current382 artifacts. Filling only `audit/<stem>.py` does not finish the383 close. Re-run step 1 until `action` is `skip`. Missing384 jupytext / nbformat / nbconvert → one-line skip naming385 `add-python-package`; do not fail the turn. `skip` /386 `not_evaluated` means the report is missing: do not fill an387 unfitted script. When `action` is `convert`, the388 unfitted-snapshot ban does not apply, including when389 `experiments/<stem>.py` already contains `skore.evaluate`.3903. record-outcome, only once step 1 is `skip`. Load391 `manage-ml-backlog` only if392 `status.skills.manage-ml-backlog` is true, in **record-outcome393 mode**, handing it the locator evaluate already returned,394 plus the optional headline and G-AUDIT-FINDING395 (`n/a — audit not run` when skipped). Else one-line skip; do396 not write History from this meta. It writes the397 `journal/JOURNAL.md` History row and design-note Status block398 plus `## Results`, then returns. It does not triage idea399 files. Never mark `done` while `smoke run` is `stop`. Missing400 headline becomes `n/a`, never an invented metric.4014. `site build` only when `policy.site` is true and402 `export-ml-site` is installed:403 `python -m skore_skills site build --if-stale`, so the Method DataOp404 report and Results replace the construct-time snapshot. Skip405 in one line otherwise. Site build embeds406 `audit/<stem>.nb.html` under `## Notebooks`; do not add407 `<!-- results-embed: audit -->`. If `site build` errors with408 `mkdocs-material is required`, load `add-python-package` for409 `mkdocs-material` (agent) and build once more. Do not410 `pixi add` / `uv add`. Name a build error; do not fail the411 turn.4125. User-facing close. 2–6 sentences of the result, grounded in413 the audit digest when present (Checks + Metrics), else the414 user's headline / `report.txt`. Do not invent a metric. When415 `policy.site` is true and the build ran, say the website has416 been updated and paste the `file://` line from that output417 verbatim. Do not name418 `html/<stem>.html` or419 `journal/<stem>.md`. When the site was not built, do not420 claim it was updated and do not name the markdown. Then421 G-REPORT-LOCATOR (the locator evaluate passed up, or422 `n/a — backend did not expose a locator`) first among423 tokens, then G-AUDIT-FINDING. Index strings, not the narrative.4246. `python -m skore_skills git end-turn --stage implement`. If425 JSON `action` is `invoke`, load `persist-ml-git` only if426 `status.skills.persist-ml-git` is true and stop; that skill427 returns to triage. If persist is missing, name the pending428 `staged` paths and stop. Otherwise load `triage-ml-task` only429 if `status.skills.triage-ml-task` is true; else stop. Do not430 run `git commit` in this skill.