| name | icfes-item-workflow |
| description | Build, critique, revise, and export ICFES-style item packages from grounded content sources for docentes/evaluadores. |
ICFES Item Workflow Skill
Use this skill when user wants to create, improve, review, batch-generate, or export ICFES-style single-answer multiple-choice items.
Inputs
Required:
- Content source: lecture, reading, existing item, rubric, image, table, graph, or equivalent material.
- Item goal or target, unless clearly inferable from source.
Optional:
- Desired number of items.
- Target difficulty.
- Area/asignatura.
- Competency/evidence.
- Classroom vs formal Saber/ICFES bank use.
If no content source exists, stop and ask for one.
Workflow
-
Inspect sources first.
- Extract source refs by file, page/slide/section if available.
- For weak extraction or visual interpretation, create
source_summary and set lower confidence.
-
Infer item targets.
- Identify area, competency, evidence, source content, target difficulty.
- List grounded assumptions.
-
Ask blocking questions only if needed.
- Group up to 3 questions for batch work.
- Do not ask pre-scan questionnaires.
-
Draft in batch waves.
- Default wave size: 5 items.
- For requests above 5, complete waves of 5 and summarize after each wave.
-
Run self-critique pass.
- Apply all quality gates.
- Revise draft before showing user.
- Keep warnings visible.
-
Output Review Markdown first.
- Use
templates/item-package.md.
- Spanish prose and labels.
- English snake_case keys where machine fields appear.
-
Export Author JSON when requested.
- Use
templates/item-package.schema.json.
- Export array of item packages.
- Include
student_view without clave, rationales, quality gates, or revision notes.
Quality Gates
source_grounding
evidence_alignment
context_sufficiency
enunciado_clarity
option_quality
difficulty_fit
An item can be ready only when all gates pass. Otherwise status is draft. Draft export with warnings is allowed.
Output Contract
Each item package must include:
item_id
status
source_refs
source_summaries
grounded_assumptions
alignment
difficulty
context
enunciado
options
clave
distractor_rationales
quality_gates
warnings
revision_notes
student_view
Review Stance
Be strict. Flag unsupported claims, ambiguous enunciados, weak distractors, multiple plausible claves, cueing, irrelevant context, and difficulty mismatch.