slxd-coursekit 0.1.2__py3-none-any.whl

coursekit/__init__.py

@@ -0,0 +1,8 @@
+"""coursekit: AI-assisted e-learning course production."""
+
+from importlib.metadata import PackageNotFoundError, version
+
+try:
+    __version__ = version("slxd-coursekit")
+except PackageNotFoundError:  # running from a source checkout without installing
+    __version__ = "0.0.0"

coursekit/__main__.py

@@ -0,0 +1,5 @@
+import sys
+
+from coursekit.cli import main
+
+sys.exit(main())

coursekit/agentkit/agents/reviewer.md

@@ -0,0 +1,29 @@
+---
+description: AI reviewer of course units. Reviews content.md and assessment.md of one unit with the content-review skill and writes the report in reviews/.
+mode: primary
+model: {{model}}
+temperature: 0.2
+permission:
+  edit: allow
+  bash:
+    "*": ask
+    "coursekit verify*": allow
+    "coursekit status*": allow
+    "coursekit brief*": allow
+    "coursekit outline*": allow
+    "coursekit config*": allow
+    "coursekit reviewed*": allow
+    "git status*": allow
+    "git diff*": allow
+    "git commit*": deny
+    "git push*": deny
+    "*coursekit approve*": deny
+    "*coursekit client*": deny
+    "*coursekit hold*": deny
+    "*coursekit resume*": deny
+    "*coursekit handoff*": deny
+    "*coursekit reviewed*--by*": deny
+---
+
+You are the e-learning content reviewer of the {{project_name}} team. Follow the `/review-unit` command
+and the `content-review` skill to the letter.

coursekit/agentkit/agents/writer.md

@@ -0,0 +1,26 @@
+---
+description: Writer of course units. Writes content.md and assessment.md of one unit from the signed design, following /write-unit and the content-writing skill.
+mode: primary
+model: {{model}}
+temperature: 0.4
+permission:
+  edit: allow
+  bash:
+    "*": ask
+    "coursekit verify*": allow
+    "coursekit status*": allow
+    "coursekit brief*": allow
+    "coursekit outline*": allow
+    "coursekit config*": allow
+    "git status*": allow
+    "git diff*": allow
+    "git commit*": deny
+    "git push*": deny
+    "*coursekit approve*": deny
+    "*coursekit client*": deny
+    "*coursekit hold*": deny
+    "*coursekit resume*": deny
+---
+
+You are the e-learning content writer of the {{project_name}} team. Follow the `/write-unit` command
+and the `content-writing` skill to the letter.

coursekit/agentkit/commands/approve-design.md

@@ -0,0 +1,19 @@
+---
+description: Prepare the sign-off of a course's instructional design (refresh the export from slxd) and tell the person how to sign it.
+argument-hint: "<CODE>"
+---
+
+Prepare the approval of the instructional design of course $ARGUMENTS.
+
+1. Read the matrix with `design_matrix_get` and compare it with `courses/<CODE>/design/matrix.json`.
+   If the person edited in slxd since the last export, export again (Excel, `matrix.json`,
+   `validation.json`).
+2. `design_matrix_validate`: if there are errors, stop and explain them.
+3. Run `coursekit sync <CODE> --check` and summarise what will be signed: units, hours, sections,
+   objectives and minimum words, plus the warnings.
+4. Tell the person that, if they agree, they sign it themselves by typing in this same chat:
+
+   `! coursekit approve design <CODE> --yes`
+
+   (or `coursekit approve design <CODE>` in their terminal). **You cannot run that command**: the
+   signature belongs to the reviewer.

coursekit/agentkit/commands/approve-unit.md

@@ -0,0 +1,18 @@
+---
+description: Check that a unit is ready for the editorial sign-off and tell the person how to sign it.
+argument-hint: "<CODE> <N>"
+---
+
+Prepare the editorial approval of the unit: $ARGUMENTS (course code and unit number).
+
+1. `coursekit verify <CODE> --unit <N>` without errors.
+2. How the unit was reviewed (`course.yaml › units[N].review.kind`): `ai` (or missing) means that
+   `courses/<CODE>/reviews/unit-NN-ai-review.md` exists and has no pending blocking findings; `human` means that a
+   person recorded their review (`by`, `note`, `report`: read them); `skipped` means that the project does not use AI
+   review. Say which one it is in the summary.
+3. Summarise the state (words, directives, placeholders, proposals awaiting a decision).
+4. If all is well, tell the person to sign it themselves:
+
+   `! coursekit approve content <CODE> --unit <N> --yes`
+
+   **You cannot run that command.**

coursekit/agentkit/commands/assemble.md

@@ -0,0 +1,10 @@
+---
+description: Assemble a course or one unit: load it into slxd creator (only what changed) or build its SCORM package with the html backend.
+argument-hint: "<CODE> [N]"
+---
+
+Assembly of: $ARGUMENTS (course code and, optionally, the unit).
+
+The project assembles with the **{{backend}}** backend by default; `coursekit status <CODE>` says the one of the course (a course
+may use the other). With `creator` load the `creator-assembly` skill; with `html` load the `html-assembly` skill. Follow it.
+Without a unit number, process every unit in order.

coursekit/agentkit/commands/client-feedback.md

@@ -0,0 +1,9 @@
+---
+description: Read the client's review comments of a course, apply the ones that make sense and answer them in creator.
+argument-hint: "<CODE> [N]"
+---
+
+Client feedback of: $ARGUMENTS (course code and, optionally, one unit).
+
+Load the `client-review` skill and follow it. Without a unit number, process every unit that has a
+review link. Never run `coursekit client`: the person records what the client decided.

coursekit/agentkit/commands/course-status.md

@@ -0,0 +1,7 @@
+---
+description: Show the state of the courses, or of one course, and the next step.
+argument-hint: "[CODE]"
+---
+
+Run `coursekit status $ARGUMENTS` and explain the result in two or three lines, in
+{{ui_language_name}}, including the next step.

coursekit/agentkit/commands/define-theme.md

@@ -0,0 +1,10 @@
+---
+description: Define the theme of the project (or of one course) in the authoring platform and derive the design tokens for the media.
+argument-hint: "[CODE]"
+---
+
+Theme definition: $ARGUMENTS (without a course code, the theme of the project; with one, the own theme of that course).
+
+Load the `theme-definition` skill and follow it (it reads the brand material in `theme/branding/` and the project's choice,
+`project.yaml › theme › source`). Never write or edit `tokens.json` by hand: it is derived from the
+theme of the platform with `coursekit theme import`.

coursekit/agentkit/commands/deliver.md

@@ -0,0 +1,11 @@
+---
+description: Export the SCORM packages of a course from slxd, record them in the course.
+argument-hint: "<CODE> [version]"
+---
+
+Delivery of the course: $ARGUMENTS (code and, optionally, version).
+
+Load the `delivery` skill and follow it unit by unit. First run `coursekit delivery check <CODE>`: if it
+refuses, stop and tell the person why. With the html backend (`coursekit status <CODE>`) there is no export in a platform: the
+steps for each unit are `coursekit assemble build <CODE> --unit N --version X.Y` and then `coursekit delivery add` without
+`--job` or `--snapshot`; skip steps 1 to 4 below.

coursekit/agentkit/commands/design-change.md

@@ -0,0 +1,14 @@
+---
+description: Apply in slxd the changes the person asks for on a course's instructional design proposal.
+argument-hint: "<CODE> <changes>"
+---
+
+Changes asked for on the instructional design: $ARGUMENTS
+
+The first argument is the course code; the rest, the changes. Run `coursekit brief <CODE>` first
+in case there is new material. Load the `instructional-design` skill and follow its **Changes**
+section.
+
+If the design was already signed (`course.yaml › approvals`), say that the change invalidates the
+signature and that it must be approved again. Summarise what changed and its consequences (hours,
+minimum words, activities). Do not sign anything and do not commit.

coursekit/agentkit/commands/new-course.md

@@ -0,0 +1,29 @@
+---
+description: Create a course from its title and hours and build the instructional design proposal in slxd for human review.
+argument-hint: "\"<title>\" <hours> [--code ABC101] [--no-intro] [--no-summary] [--no-material] [indications]"
+---
+
+Create a new course and its instructional design **proposal**. Arguments from the person:
+
+$ARGUMENTS
+
+1. Separate the title (in quotes), the hours, the options `--code`, `--no-intro`, `--no-summary`,
+   `--no-material` (only for you: do not pass it to `coursekit new`) and the rest as free indications. If the title or the hours are missing, ask for them and stop.
+2. Run `coursekit new "<title>" <hours> [options] --notes "<indications>"`, unless `courses/<CODE>/course.yaml` already
+   exists (the course was created before, for example in handoff mode): then skip this step and work on it as it is.
+3. Run `coursekit brief <CODE>` and read `brief/index.md` and `brief/notes.md`. If there is no
+   material (no documents, no web links, no notes beyond the template, in the course or in the
+   project) and `--no-material` was not given, **stop before designing** and ask the person, in
+   {{ui_language_name}}: they can drop files in `courses/<CODE>/brief/sources/` (or the project's
+   `brief/sources/`), URLs in `brief/links.md` and indications in `brief/notes.md`, and say when
+   it is there (then run `coursekit brief <CODE>` again), or tell you to go on without it. If you
+   cannot ask (non-interactive session) or the person goes on without material, say in the final
+   summary that the syllabus is an assumption and that `/design-change` redoes it with material.
+4. Load the `instructional-design` skill and follow its **Proposal** section.
+5. Run `coursekit theme show`. If the project has no theme tokens, say in the summary that `/define-theme`
+   must be run before producing media.
+6. Finish with a short summary for the person, in {{ui_language_name}}: units with hours and
+   objectives, decisions and assumptions to check, where the Excel is, and how to go on: changes
+   with `/design-change <CODE> <changes>` and sign-off with `/approve-design <CODE>`.
+
+Do not sign anything and do not commit.

coursekit/agentkit/commands/produce-media.md

@@ -0,0 +1,10 @@
+---
+description: Produce the pending media assets of a course with the available providers and upload them.
+argument-hint: "<CODE> [asset id]"
+---
+
+Media production of: $ARGUMENTS (course code and, optionally, one asset).
+
+Load the `media-production` skill and follow it. Start with `coursekit media extract` and
+`coursekit media plan`, and tell the person what you will produce and with what before spending
+credits of paid APIs.

coursekit/agentkit/commands/review-unit.md

@@ -0,0 +1,12 @@
+---
+description: AI review of a written unit (reviewer agent of REVIEWER_AGENT / REVIEWER_MODEL), with a report in reviews/.
+argument-hint: "<CODE> <N>"
+agent: reviewer
+---
+
+Review the unit: $ARGUMENTS (course code and unit number).
+
+Load the `content-review` skill and follow it. If the message says PARTIAL REVIEW, follow its
+"Partial review" section. The review should be done by a model other than the writer's: check
+`written_with` of the unit in `courses/<CODE>/course.yaml`; if it matches the tool and model you
+are running on, say so at the start and in the report, and go on with the review.

coursekit/agentkit/commands/sync-directives.md

@@ -0,0 +1,19 @@
+---
+description: Sync the directive registry of the project with the slxd creator brick catalog (list_brick_types).
+---
+
+Update the directive registry with the current creator catalog.
+
+1. Call `list_brick_types` (slxd MCP) and save the JSON as it is in `.cache/list_brick_types.json`.
+2. Run `coursekit directives check .cache/list_brick_types.json`.
+3. If it says `ok`, update only `synced_with_creator` with today's date in
+   `config/directives.yaml` (create it with that key if missing) and finish.
+4. For each `NEW`: call `get_brick_type_schema` of the type and decide whether it should be a
+   directive.
+   - If so: add it to `directives` in `config/directives.yaml` with a kebab-case name, `brick`,
+     `role` (`interactive`, `question` or `static`) and `use` in {{ui_language_name}}.
+   - If not: add it to `not_directives` with the reason.
+5. For each `REMOVED`: remove it from the registry. If it was a directive, find its uses in
+   `courses/*/content/**/*.md` and list the affected files (do not change them).
+6. Update `synced_with_creator`, run step 2 again until it says `ok` and summarise the changes.
+   Do not commit.

coursekit/agentkit/commands/write-unit.md

@@ -0,0 +1,25 @@
+---
+description: Write one unit of a course with a signed design, or every unit still to write (writer agent of WRITER_AGENT / WRITER_MODEL).
+argument-hint: "<CODE> [N]"
+agent: writer
+---
+
+Write: $ARGUMENTS (course code and, optionally, a unit number). If the code is missing, ask.
+
+Load the `content-writing` skill and follow it. Before writing anything, check with
+`coursekit status <CODE>` that the instructional design is signed; if not, stop and say so.
+
+- **With a unit number**, write that unit only.
+- **Without a number**, write every unit still to write, in order: the ones whose status in
+  `courses/<CODE>/course.yaml › units[].status` is `pending` or `writing`. Finish a unit completely
+  (`coursekit verify <CODE> --unit N` passes and it is `verified`) before starting the next, and stop at
+  the first one that does not verify, saying why. Between units give a one-line summary and go on without
+  asking. Do not re-read the units you finished: for continuity use `coursekit outline <CODE>` and
+  `coursekit outline <CODE> --section U.S`, as the skill says.
+- You work on one unit at a time: `courses/<CODE>/content/unit-NN/`. Do not touch the other units, nor
+  `config/`, nor `project.yaml`.
+- Write section by section and run `coursekit verify <CODE> --unit N` after each one. Do not go
+  on with errors.
+- {{language_name}}, {{address_rule}}, no emojis.
+- If a fact cannot be verified, mark it with `<!-- VERIFICAR: … -->`; do not invent it.
+- Do not commit or push, and never run `coursekit approve`.

coursekit/agentkit/docs/content-format.md

@@ -0,0 +1,126 @@
+# Content format — `content.md` and `assessment.md`
+
+Canonical format of a unit's content. The writing agent writes it, `coursekit verify` checks it
+and the assembly turns it into the platform's blocks. Course language: {{language_name}}; the
+format words below (section heading, objective tag, media placeholder) are in that language.
+
+## 1. Structure of `content.md`
+
+```markdown
+# {{t_unit}} N — <unit title>
+
+> unit header: duration, minimum pages and words, unit objectives (UN.1, UN.2…)
+
+---
+
+## {{t_section}} 1 — {{t_intro_title}} *(…)*
+### Subsection
+…
+## {{t_section}} 2 — <title> *(…)*
+…
+*[{{t_objective}} UN.1 — Verb]*
+…
+## {{t_section}} N — {{t_summary_title}} *(…)*
+```
+
+- Each section opens with `## {{t_section}} N — <title> *(minimum)*`. Number, order, title and
+  minimum come from the approved instructional design (`course.yaml › units › sections`): the
+  minimum is {{words_per_hour}} words per hour of the section ({{pages_per_hour}} pages of
+  {{words_per_page}} words).
+- Each section becomes one lesson.
+- Subsections with `###`. Prose, lists, Markdown tables and code blocks are converted as they are.
+- Code blocks: write ```` ```bash ```` for commands (also PowerShell and `cmd`) and ```` ```text ```` for what a program prints or a
+  conversation. Creator has no plain-text language, so a `text` block made only of commands is shown as `bash` and the rest
+  uses the platform's automatic detection, which guesses badly with short code: keep outputs in `text` and commands in `bash`.
+- The tag `*[{{t_objective}} UN.M — Verb]*`, one per line, closes each content section. It is
+  editorial metadata: it is not published.
+
+**Word count:** only text the learner reads counts. Media placeholders, objective tags, HTML
+comments and the `:::` lines of directives do not count (the text inside a directive does).
+
+Writing rules: {{language_name}}, {{address_rule}}, complete prose (no bare lists or unexpanded
+outlines), **no emojis** (only `{{allowed_symbols}}` are allowed).
+
+## 2. Interactive component directives
+
+A component is written as a `:::name` … `:::` block and becomes one block of the assembly
+platform. The full registry of directives, their block and when to use each one is
+`coursekit config directives` (`directives` and `use`): it is the only valid list. Common
+components that are not directives have their equivalent in `component_equivalents`.
+
+Each content section includes **distinct** interactive directives in proportion to its hours
+({{interactive_per_hour}} per hour, at least {{min_interactive_per_content_section}}). `note`,
+`highlight` and `quote` do not count as interactive.
+
+Internal structure (anything else fails at assembly):
+
+| Directive | Internal structure |
+|---|---|
+| `accordion`, `tabs`, `carousel` | one `#### Title` per panel, tab or slide + Markdown text |
+| `timeline` | `#### <date> — <title>` per milestone (without " — ", all of it is the title) + text |
+| `flashcards`, `flashcard-gallery` | `#### Front` + text of the back |
+| `labelled-graphic` | `{{k_image}}: <exact title of the placeholder, as in `media/manifest.yaml`>` + `#### Point` + text; optional `{{k_position}}: x,y` (0–100) under each point |
+| `dialog` | `**Character:** line`, one per line |
+| `carousel-quotes` | one quote per paragraph, with `— author` on its last line |
+| `note` | optional `{{k_title}}: …` + text |
+| `highlight` | text |
+| `quote` | quote + `— author` on the last line |
+| `word-search`, `wordle`, `hangman` | `- WORD` per line |
+| `memory` | `- card content` per line |
+| `pasapalabra` | `- A ({{w_starts}}): definition :: ANSWER` or `({{w_contains}})`; alternatives with `/` |
+| `trivial` | `#### Category` + `{{k_question}}: …` blocks with `- [x]` / `- [ ]` options, separated by a blank line |
+
+A plain blockquote (`> text`) becomes a highlight.
+
+## 3. Question directives (practice activities and tests)
+
+The directives with `role: question`: `single-choice`, `multi-select`, `true-false`, `sorting`,
+`match`, `sorting-groups`, `fill-in-the-blank`, `order-words`, `short-answer`.
+
+```markdown
+:::single-choice
+{{t_question_key}} U1.2
+{{k_question}}: <question>
+- [ ] <distractor>
+- [x] <correct answer>
+- [ ] <distractor>
+{{k_feedback_correct}}: <why it is right>
+{{k_feedback_incorrect}}: <which section to review>
+:::
+```
+
+- `single-choice` (exactly one `- [x]`) and `multi-select`: options `- [x]` / `- [ ]`.
+- `match`: lines `- term :: definition`. `sorting`: list in the right order. `order-words`:
+  `- full sentence` per sentence. `sorting-groups`: `#### Category` + list of items.
+  `true-false`: `{{k_answer}}: {{w_true}}|{{w_false}}`. `fill-in-the-blank`: blanks in `{{k_question}}:` as
+  `{{{k_answer}}}` (several accepted: `{a/b}`). `short-answer`: `{{k_answers}}: a | b | c`.
+- Common keys: `{{t_question_key}}`, `{{k_question}}:`, `{{k_feedback_correct}}:`, `{{k_feedback_incorrect}}:`,
+  `{{k_feedback}}:`. Each key on one line. Every question carries `{{t_question_key}}` and feedback.
+
+## 4. Media placeholders
+
+Exact format (shown as a visible box until the asset is produced):
+
+```markdown
+> **[{{t_placeholder}} — <type>]**
+{{placeholder_field_lines}}
+```
+
+Allowed types (what is written in the placeholder, and its id in the configuration): {{placeholder_types}}. Per unit: {{placeholders_per_hour}} placeholders per hour
+of the unit and at least {{min_placeholder_types}} different types.
+
+## 5. `assessment.md`
+
+The summative activities of the unit, in order:
+
+```markdown
+## {{t_assessment}} N.M — <name>
+### (objectives assessed · instructions for the learner · development · automatic marking and
+###  scoring · attempts and feedback · question bank)
+:::single-choice
+…
+:::
+```
+
+The unit test bank has {{questions_per_objective}} questions per objective assessed by
+questionnaire.

coursekit/agentkit/docs/slxd-mcp.md

@@ -0,0 +1,41 @@
+# slxd MCP (creator) — reference for coursekit
+
+> Before using a tool, confirm its schema with `tool_schema` (or `get_brick_type_schema` for
+> bricks): this page is a guide, not the contract.
+
+## Connection
+
+- The project server is **`{{mcp_name}}`** (`project.yaml › platform.slxd`), configured in
+  `.mcp.json`, `opencode.json` and `.codex/config.toml` by `coursekit agents`. Each person
+  authenticates with OAuth (slxd account) the first time.
+- Streamable HTTP transport; **OAuth 2.1** or an `sk_…` key.
+- An MCP credential always acts as **member**: owner/admin actions are not available.
+- Destructive tools need `confirm: true`; without it they return a preview.
+- **Direct tools** include `browse_workspace`, `get_content`, `get_lesson`, `create_content`,
+  `update_content`, `create_lesson`, `update_lesson`, `reorder_lessons`, `delete_lesson`,
+  `list_brick_types`, `get_brick_type_schema`, `add_brick`, `update_brick`, `move_brick`,
+  `delete_brick`, `audit_content_accessibility`, `apply_accessibility_autofix`, `list_themes`,
+  `get_theme`, `update_theme`, `request_asset_upload`, `search_stock_images`, `create_export`,
+  `get_export_status` and `design_matrix_create/get/list/validate/generate_course`.
+- **The rest** (other `design_matrix_*`, `design_matrix_export`, `create_snapshot`,
+  `share_content`, `update_quiz_settings`, `set_content_theme`, `create_theme`,
+  `import_stock_image`, `search_stock_icons`, `request_embed_upload`…) go through
+  **`find_tools`** → **`tool_schema`** → **`run_tool`**.
+
+## Tools by phase
+
+| Phase | Tools |
+|---|---|
+| Theme | `list_themes`, `create_theme`, `get_theme`, `update_theme`, `set_content_theme` |
+| Design | `design_matrix_*` (matrix, competencies, objectives, nodes, learning and assessment activities, validate, export) |
+| Skeleton | `design_matrix_generate_course`, `design_matrix_redaccion_status` |
+| Media | `request_asset_upload`, `request_embed_upload`, `search_stock_images`, `import_stock_image`, `search_stock_icons`, `import_stock_icon` |
+| Assembly | lessons and bricks, `update_quiz_settings`, `set_lesson_bank_references`, accessibility audit, `create_snapshot` |
+| Review | `share_content` (live preview) |
+| Delivery | `create_export`, `get_export_status` |
+
+Useful values:
+- `design_matrix_create.patronEstructural`: `con_modulos | sin_modulos` (cannot be changed).
+- `nivelBloom`: `recordar | comprender | aplicar | analizar | evaluar | crear`.
+- Assessment `instrumento`: `cuestionario | caso_practico | entrega_portfolio | rubrica`.
+- `create_export.standard`: `scorm_1_2 | scorm_2004`.

coursekit/agentkit/skills/_slxd-tools.md

@@ -0,0 +1,10 @@
+## Calling the slxd tools
+
+The MCP server `{{mcp_name}}` lists only its main tools directly. **If a tool named in this skill
+is not in your list** (e.g. `design_matrix_competencia_create`, `design_matrix_export`,
+`create_snapshot`, `share_content`, `enable_review`, `update_quiz_settings`), do not assume it does not exist:
+1. `find_tools` with keywords (e.g. `"design_matrix objective"`) to get its exact name;
+2. `tool_schema` with that name to see its arguments;
+3. `run_tool` with `name` and `arguments`.
+Direct tools are called as usual. Destructive tools (`confirm: true`) are confirmed with the
+person first.

coursekit/agentkit/skills/client-review/SKILL.md

@@ -0,0 +1,50 @@
+---
+name: client-review
+description: Comments of the client on the review link of a course in slxd creator — read them, apply the agreed changes in the .md, reload, publish a new review version and answer each comment. Use with /client-feedback or when the client has commented.
+---
+
+# The client's review
+
+The review of each unit is a link of its creator content (`course.yaml › units[].links.review`); the client
+comments on it without an account. A round is opened and closed by the person with `coursekit client`.
+The `.md` rules still apply: never edit content directly in creator; fix the `.md` and reload it.
+
+{{> _slxd-tools}}
+
+## 1. Read the comments
+
+For each unit with a `content_id` and a review link:
+1. `get_review`: enabled, latest version and how many threads are pending.
+2. `list_review_comments` with `status: pending` (use `since` to read only what is new). Each thread says the
+   lesson (`lessonTitle`), the text and, if any, a screenshot (`screenshotUrl`: open it).
+3. Map each comment to the section or activity of the `.md` (`coursekit outline <CODE> --section …`).
+
+## 2. Decide with the person
+
+Show a table per unit: comment, section, what you propose (apply, do not apply and why, needs the person).
+In an interactive session wait for the person's decision; without an interface, apply only the clear text
+corrections that fit the design and leave the rest in the summary. Comments that change objectives, hours,
+activities or the structure are a design change: do not apply them, point to `/design-change`.
+
+## 3. Apply
+
+1. Edit the `.md` with the `content-writing` rules (language, tone, no emojis) and run `coursekit verify <CODE>`.
+2. A unit edited after its signature goes back to `verified`: tell the person it needs a new review and signature
+   (`/review-unit`, `/approve-unit`) before the course is delivered.
+3. Reload what changed: `coursekit assemble plan`, `diff` and apply as in the `creator-assembly` skill.
+
+## 4. Answer
+
+For each unit that changed:
+1. `publish_review_version` with a label (`v1.1`, `v1.2`…): the client sees the update. Only the latest version
+   accepts comments.
+2. For each applied comment: `reply_review_comment` on the root of the thread, saying what changed (and for the
+   ones you did not apply, why), then `set_review_comment_status` with `resolved`. `verified` is for the client.
+3. When the person closes the round, `export_review_comments` and download the CSV (the link lasts 24 hours) to
+   `.cache/client-review/<CODE>-round-N.csv` (outside git and never published); it has the e-mail addresses of guests: do not paste it anywhere nor move it into the course.
+
+## 5. Summary
+
+Comments read, applied, discarded and pending; units reloaded and the new review version of each; what the
+person has to do next (`coursekit client <CODE> send` for another round, or `approve` when the client agrees).
+Do not commit unless asked.

coursekit/agentkit/skills/content-review/SKILL.md

@@ -0,0 +1,78 @@
+---
+name: content-review
+description: Quality review (second layer, by the reviewer agent of the .env, other than the writer) of a written unit. Use to review content.md/assessment.md of courses/<CODE> and write reviews/unit-NN-ai-review.md.
+---
+
+# AI review of a unit
+
+The reviewer should **not** be the model that wrote the unit: compare with `written_with` of the
+unit in `course.yaml`; if it matches, say so in the report and carry on.
+
+## Preparation
+
+1. `coursekit verify <CODE> --unit N`: errors are fixed first (they are objective and cheap).
+2. Read `course.yaml` (objectives, sections and activities of the unit, audience, level), the
+   design Excel, `design/proposal-notes.md` and the whole unit.
+3. Reference material: `coursekit brief <CODE>`, then `brief/index.md` and `brief/notes.md` of the
+   course and of the project. Check technical accuracy (criterion 1) against `brief/text/` and
+   resolve the `<!-- VERIFICAR -->` with it; anything contradicting the material is **blocking**.
+
+## Rubric
+
+Assess each section with these criteria. Every finding has a severity: **blocking** (cannot
+reach the learner), **improvement** (noticeably better quality) or **minor** (style, typos).
+
+| # | Criterion | What to check |
+|---|---|---|
+| 1 | Technical accuracy | Facts, versions, commands, prices, regulations and names correct and current. Find and resolve the `<!-- VERIFICAR -->`. |
+| 2 | Alignment | Each content section lets the learner reach its objective at the declared Bloom level; activities practise and assess **that** objective. |
+| 3 | Coherence with the design | Titles, sections and activities match the matrix; nothing out of scope. |
+| 4 | Depth | Developed prose, no padding or repetition to reach the minimum; concrete professional examples. |
+| 5 | Activities | Match the design; unambiguous questions, plausible distractors, one right answer in single-choice, useful feedback pointing to the section. {{questions_per_objective}} questions per objective assessed by questionnaire. |
+| 6 | Interactivity | Directive suited to the kind of content; no flat section. |
+| 7 | Media | Necessary placeholders, well specified, with accessibility (captions, alt text). |
+| 8 | Style | {{language_name}}, {{address_rule}}, tone of `course.yaml`, inclusive language, no emojis. |
+| 9 | Continuity | No overlap with other units nor content the design keeps for a later one (`coursekit outline <CODE>`; to check an overlap, `--section U.S`); correct cross-references. |
+
+## Partial review
+
+When the unit was already reviewed and only some parts changed (`coursekit review` computes
+them against the previous review, or the person names them with `--parts`):
+
+1. Apply the rubric **only** to those parts (sections of `content.md`, activities of
+   `assessment.md`). Of the rest, read only what you need to check how they fit: the objective
+   they develop, the neighbouring sections (continuity, cross-references) and the bank questions
+   and activities of the same objective. Use `coursekit outline <CODE>` and `--section U.S`
+   instead of reading the whole unit.
+2. Do not reopen findings already resolved in parts that did not change.
+3. Do not write a new report: **append** to `reviews/unit-NN-ai-review.md` a section
+   `## Partial review — YYYY-MM-DD (<model>)` with the parts reviewed, the findings table and the
+   changes applied, and update the **Result** in the header if it changes.
+4. Finish as usual: `coursekit reviewed <CODE> <N>`.
+
+## Output
+
+1. **Apply directly** the blocking and minor fixes in the `.md`. Leave the improvements that
+   change the approach as proposals in the report.
+2. Write `courses/<CODE>/reviews/unit-NN-ai-review.md` (in {{ui_language_name}}):
+
+```markdown
+# AI review — <CODE> · Unit N
+> **Reviewer:** <model> · **Date:** YYYY-MM-DD · **Result:** ready | ready with changes | not ready
+<!-- result: ready | ready_with_changes | not_ready -->
+
+## Summary
+## Findings
+| # | Section | Criterion | Severity | Finding | Action (applied / proposed) |
+## Changes applied
+## Proposals awaiting a human decision
+```
+
+   The `<!-- result: … -->` line repeats the **Result** with one of its three values, as written there (untranslated,
+   with underscores): tools read it. Update it too when a partial review changes the Result.
+3. Run `coursekit reviewed <CODE> <N>`: it verifies again and marks the unit as reviewed (if it
+   fails, fix and repeat). Never edit statuses by hand. Never pass `--by` to it: that option records the review
+   of a person.
+4. Remind the person that, after their editorial review, they sign with `/approve-unit <CODE> <N>`.
+
+Do not commit or push unless asked.