writing-evaluator

Writing Evaluator

Generates original writing prompts and evaluates responses. Paths are relative to ${CLAUDE_PLUGIN_ROOT} (if unset, resolve relative to the plugin root — the directory two levels up from this file, i.e. the folder containing data/ and skills/).

When to use

The user pasted a piece of writing for assessment, OR asked for a practice writing task in a named exam format, OR asked how to improve exam writing.

Steps

1. Identify exam + level + task type. If unclear, ask once. When EVALUATING pasted writing, also obtain the exact task prompt as it was set — the IELTS question, the two given points for a B2 essay, the TOEFL email's three content points, the letter's bullets — plus whether it was written timed and in how long. Task fulfilment (were the required points covered, right genre, word count) is the first-ranked criterion and caps the band; without the prompt, say so and mark Task Achievement provisionally rather than guessing. Load:

   • data/exam-formats/<exam-id>.md (format, word count, timing, criteria names)
   • data/cefr/writing-descriptors.md (assessment anchors)
   • data/cefr/calibration-anchors.md (leveled reference samples)
   • skills/writing-evaluator/references/task-anatomy.md (structure and register expectations per task type)

2. If GENERATING a prompt: produce an ORIGINAL prompt matching the format's structure, word count and timing (seed examples: data/item-bank/seed/writing-prompts.md — imitate shape, never reuse content). Pitch the prompt's cognitive/topic demand to the level, not just its length: B1 concrete/personal; B2 a familiar issue to take a stance on; C1 abstract, requiring evaluation/weighing; C2 nuanced or counter-intuitive. (calibration-anchors.md shows what an at-level answer looks like — aim the prompt so a level-appropriate answer lands there.) State the time limit and word target. Offer to time the attempt: note the start time, and compute elapsed seconds when the answer arrives.