Use prompt workflows
The server exposes nine deterministic MCP prompt workflows. A prompt does not call an LLM, retrieve evidence by itself, or generate the final educational material. It renders versioned instructions and runtime context for the connected MCP host, which then calls the appropriate tools and produces any final synthesis or draft.
This separation lets the server enforce framework identity, rights, attribution, retrieval boundaries, and required disclosures without presenting model-generated text as source-authored curriculum.
Prompt execution model
%%{init: {"themeVariables": {"fontSize": "18px"}}}%%
flowchart LR
A[Prompt arguments] --> B[Resolve exact package and profile]
B --> C[Rights and capability checks]
C --> D[Merge generic and framework-local guidance]
D --> E[Rendered MCP prompt]
E --> F[MCP host]
F --> G[Tool calls for source evidence]
G --> F
F --> H[Generated answer or draft]Framework-local prompts.json files can append to or replace declared soft guidance
blocks. They cannot override correctness-critical rights, attribution, privacy,
evidence-status, package-isolation, or generated-content rules.
Available prompts
| Prompt | Primary use |
|---|---|
student_study_support |
Evidence-grounded study support and practice |
teacher_guide_draft |
Evidence-grounded teacher lesson-guide draft |
student_handbook_section |
Evidence-grounded student-facing handbook section |
multigrade_lesson_plan |
One lesson for a classroom holding several grades, built from shared learning components |
learning_progression_teaching_sequence |
Draft a cited sequence using stored links and LCs |
learning_progression_support_plan |
Propose target-standard review/practice options |
learning_progression_curriculum_review |
Inspect bounded counts, warnings and provenance |
administrator_alignment_review |
Structured review between one source and one target framework |
cross_framework_comparison |
Exploratory evidence-grounded comparison across two to eight frameworks |
All nine prompts are registered as generated-content workflows. The current prompt
version is 1.4.0.
Seven of them (all except administrator_alignment_review and
cross_framework_comparison) can also be fetched as a tool result with
get_workflow_instructions, for clients that work only through tools. The rendered
message is identical to the native prompt. See
Evidence and workflow access tools.
Rights gate
Before a generated-content workflow is rendered, the server requires the selected package or packages to have:
allowGeneratedDerivatives: allowed; and- a rights review status of
approvedorprovisional_operator_approved.
If any selected package fails those requirements, prompt rendering is denied. A local prompt configuration cannot weaken this policy.
The rendered workflow also carries the applicable attribution statement and instructs the host to repeat it in the eventual answer.
Focus modes
The role-oriented prompts use focus_mode to interpret topic_or_standard:
focus_mode |
topic_or_standard value |
|---|---|
topic |
Source-visible topic or label text |
statement_code |
Stable framework statement code |
node_id |
Exact package-local node UUID |
case_identifier_uuid |
Exact CASE UUID |
case_identifier_uri |
Exact CASE URI |
statement_code is rejected when the selected framework does not provide stable code
support. Use topic or an exact identifier instead.
student_study_support
Use this workflow to generate bounded student support tied to retrieved curriculum evidence.
| Parameter | Required | Default / bounds |
|---|---|---|
framework_id |
yes | Exact framework ID |
grade_or_stage |
yes | Local or normalized grade/stage context |
topic_or_standard |
yes | Interpreted by focus_mode |
focus_mode |
no | topic |
difficulty |
no | on_level; also foundational, extension |
practice_count |
no | 5, from 1 through 10 |
local_context |
no | Anonymous local nuance |
output_language |
no | Optional BCP 47-style language tag |
snapshot_id |
no | Unique-current routing when omitted |
Example manual prompt inputs:
difficulty: on_level
focus_mode: statement_code
framework_id: ghana-nacca-primary-english-language-basic-1-3
grade_or_stage: BASIC 1
local_context: paper, pencils, and a chalkboard
output_language: en
practice_count: 5
topic_or_standard: B1.2.7.2.6
The host should preserve the retrieved source wording and clearly distinguish generated study explanations or practice items from official curriculum text.
teacher_guide_draft
Use this workflow when the desired output is a teacher-facing lesson or activity guide.
| Parameter | Required | Default / bounds |
|---|---|---|
framework_id |
yes | Exact framework ID |
grade_or_stage |
yes | Local or normalized grade/stage context |
topic_or_standard |
yes | Interpreted by focus_mode |
focus_mode |
no | topic |
lesson_duration_minutes |
no | 45, from 10 through 240 |
available_materials |
no | Untrusted material constraints |
learner_context |
no | Anonymous learner context; no sensitive education records |
local_context |
no | Untrusted local nuance |
output_language |
no | Optional language tag |
snapshot_id |
no | Unique-current routing when omitted |
Example:
available_materials: pictures and chalkboard
focus_mode: node_id
framework_id: ghana-nacca-primary-english-language-basic-1-3
grade_or_stage: BASIC 1
lesson_duration_minutes: 35
output_language: en
topic_or_standard: aa4cdccd-e5d9-589c-8094-e10d7ac0c754
Materials, local context, and learner context are caller-supplied constraints. They are not source curriculum evidence and should not be represented as such.
student_handbook_section
This workflow generates a student-facing explanatory section grounded in bounded source evidence.
| Parameter | Required | Default / bounds |
|---|---|---|
framework_id |
yes | Exact framework ID |
grade_or_stage |
yes | Local or normalized grade/stage context |
topic_or_standard |
yes | Interpreted by focus_mode |
focus_mode |
no | topic |
target_word_count |
no | 500, from 150 through 1500 |
local_context |
no | Anonymous local nuance |
output_language |
no | Optional language tag |
snapshot_id |
no | Unique-current routing when omitted |
Example:
focus_mode: statement_code
framework_id: ghana-nacca-primary-mathematics-basic-4-6
grade_or_stage: BASIC 6
output_language: en
target_word_count: 500
topic_or_standard: B6.1.4.1
multigrade_lesson_plan
Use this workflow to plan one lesson for a classroom that holds several grades at once. It separates a shared teach-together core from grade-specific work by reading which learning components the standards of each grade decompose to. A component that supports standards in more than one of the requested grades is a candidate for the shared core; a component that supports only one grade becomes that grade's differentiated work.
Experimental, and it requires learning components
This workflow has no source-only fallback. A package that contains no learning
components is refused with capability_unavailable rather than degraded into
parallel single-grade plans under a multi-grade heading.
| Parameter | Required | Default / bounds |
|---|---|---|
framework_id |
yes | Exact framework ID |
grades_in_room |
yes | JSON array of 2 through 8 distinct grades |
topic_or_standard |
yes | Interpreted by focus_mode |
focus_mode |
no | topic |
lesson_duration_minutes |
no | 45, from 10 through 240 |
learner_context |
no | Anonymous learner context; no sensitive education records |
local_context |
no | Untrusted local nuance |
output_language |
no | Optional language tag |
snapshot_id |
no | Unique-current routing when omitted |
Example:
framework_id: ghana-nacca-primary-mathematics-basic-4-6
grades_in_room: ["BASIC 4", "BASIC 5", "BASIC 6"]
lesson_duration_minutes: 40
output_language: en
topic_or_standard: fractions
The rendered workflow directs the host to resolve each grade's standards, call
get_learning_components_for_standard for each one, and compare the supported standards
across grades.
Learning components are model-generated decompositions of published standards, not source-asserted curriculum. A shared core therefore rests on a model's judgement that two standards decompose to the same component, not on a curriculum-authored equivalence between those grades, and the generated plan must say so.
Stored progression workflows
Use the three distinct prompts for different educational decisions:
| Prompt | Purpose |
|---|---|
learning_progression_teaching_sequence |
An adaptable cited sequence from stored builds, related concepts and supporting LCs |
learning_progression_support_plan |
Target-standard review/practice options using teacher observations and incoming/upstream evidence |
learning_progression_curriculum_review |
Bounded inspection of stored counts, scope, warnings and provenance for ministry/ed-tech review |
See worked prompt examples and exact prompt arguments. Each prompt pins the exact package and instructs the client to retrieve finite evidence before composing the output. Suggested pedagogy is generated; stored LP is model-generated evidence with provenance, not publisher-endorsed or certified pedagogy. No absent-edge fallback is provided.
Teacher-guide, study-support, handbook and multigrade workflows also retrieve optional stored progressions for at most three selected standards. For each standard they make three calls, one per connection kind (outgoing builds, incoming builds, related), at most nine in total. Each call reads one page of up to 25 links; a page can hold fewer if the result reaches the size limit. If a nextCursor remains, that kind has more links than were read, and the workflow says so. Asking for each kind separately makes sure related concepts are not crowded out by builds. They also read full provenance for up to ten used relationships, and supporting LCs. Related concepts stay separate from directional builds. They explain unavailable/empty/incomplete evidence and continue their useful established outputs. Shared LCs do not establish progression or equivalence.
administrator_alignment_review
This prompt renders a controlled comparison workflow for one source framework and one
target framework. The prompt itself does not retrieve standards; it instructs the host
to use compare_framework_evidence and preserve both packages' evidence boundaries.
| Parameter | Required | Default / bounds |
|---|---|---|
source_framework_id |
yes | Exact source framework ID |
target_framework_id |
yes | Exact target framework ID |
topic_or_query |
yes | Shared text or code query |
search_mode |
no | text; also code_exact, code_prefix |
source_grade_or_stage |
no | Framework-local review scope |
target_grade_or_stage |
no | Framework-local review scope |
source_snapshot_id |
no | Exact source snapshot |
target_snapshot_id |
no | Exact target snapshot |
matches_per_framework |
no | 5, from 1 through 10 |
include_context_paths |
no | true |
local_context |
no | Untrusted administrative context |
output_language |
no | Optional language tag |
Example:
include_context_paths: true
matches_per_framework: 5
output_language: en
search_mode: text
source_framework_id: nigeria-nerdc-mathematics-primary-1-3
source_grade_or_stage: PRIMARY TWO
target_framework_id: rwanda-reb-mathematics-lower-primary-1-3
target_grade_or_stage: P2
topic_or_query: length measurement
The source and target grade values are framework-local review scopes. Their presence in one workflow does not assert grade equivalence.
The eventual review should end with human-review questions before any governance, adoption, or alignment decision is made.
cross_framework_comparison
Use this prompt for exploratory synthesis across two to eight frameworks.
| Parameter | Required | Default / bounds |
|---|---|---|
framework_ids |
yes | JSON array of 2–8 distinct framework IDs |
topic_or_query |
yes | Shared text or code query |
search_mode |
no | text; also code_exact, code_prefix |
snapshot_ids |
no | JSON array, at most one snapshot per framework |
local_grade_labels |
no | Shared JSON-array local-grade filter |
normalized_grades |
no | Shared JSON-array normalized-grade filter |
matches_per_framework |
no | 5, from 1 through 10 |
include_context_paths |
no | true |
local_context |
no | Untrusted local nuance |
output_language |
no | Optional language tag |
Example:
framework_ids: ["nigeria-nerdc-mathematics-primary-1-3", "rwanda-reb-mathematics-lower-primary-1-3"]
include_context_paths: true
local_grade_labels: []
matches_per_framework: 5
normalized_grades: []
output_language: en
search_mode: text
snapshot_ids: []
topic_or_query: length measurement
Because local_grade_labels is shared across all selected frameworks, leave it empty
when the frameworks use different local labels unless the exact shared filter is
intentional.
See Compare framework evidence for the underlying deterministic tool.
Framework-local prompt guidance
Each accepted framework can have a versioned prompt configuration associated with its profile. A configuration can contribute bounded guidance such as local terminology, classroom context, warnings, and output conventions.
The rendered prompt records whether an overlay was configured and retains its prompt configuration ID, version, and SHA-256 when present. This makes the instructions used by a generated workflow auditable alongside the package and profile identity.
Soft guidance cannot redefine the source
A framework-local overlay may refine how the host works with the curriculum, but it cannot change package identity, source wording, rights, evidence status, or the server's non-overridable claim boundaries.
Prompt arguments are not source evidence
Caller-supplied values such as local_context, available_materials, and
learner_context are treated as untrusted contextual input. Keep them separate from
source-authored curriculum evidence in the final answer.
Avoid placing personally identifiable learner data, sensitive education records, or other unnecessary personal information into prompt arguments.
Client behavior
Prompt-selection UX is client-specific. Some hosts expose server prompts directly in a
prompt picker, while others may require explicit selection or manual parameter entry.
In Claude Desktop the prompts appear under Add from curriculum-knowledge-graph and
are added as an attachment. If a client does not show prompts, ask it to call
get_workflow_instructions and follow rendered.message. The server contract remains
the same: the prompt is deterministic instructions, and the host is responsible for
executing the requested tool workflow. See
Use with Claude Desktop and claude.ai.
In the seven single-framework workflows, the EVIDENCE ACCESS section explains how to read full evidence: natively
where the client can open resources, otherwise with read_evidence, finishing
multi-window records before relying on them. The EVIDENCE LINKS block gives exact
links for the pinned snapshot and patterns for standard and learning-component records.
A useful verification pattern is to inspect the rendered workflow before trusting the final generated answer. The workflow should identify the exact framework/snapshot context, source rights and attribution, required disclosures, and which retrieval tool the host must use.
Evidence and generated content
When a prompt workflow produces educational prose, keep these categories visibly separate:
| Category | Treatment |
|---|---|
| Exact source wording | Quote or reproduce only within applicable rights and attribution requirements |
| Retrieved source evidence | Preserve exact identifiers, package identity, and warnings |
| Deterministic server guidance | Describe as server-generated workflow instructions |
| Model explanation or draft | Label as generated content, not official curriculum wording |
| Comparison and stored LP | Preserve generated origin; stored LP has exact retained provenance |