Semantic consistency auditor

Skill alainlebret/claude-agents/higher-ed-teaching-agents/skills/semantic-consistency-auditor

Audits deep semantic alignment between mission, statement, rubric, reference solution, and tests. Use when artifacts may be structurally valid but pedagogically inconsistent.From its SKILL.md

Install

npx -y skills add alainlebret/claude-agents --skill semantic-consistency-auditor

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

1.7 KB, 317 tokens by cl100k_base, as published. Nobody here has run it

Semantic Consistency Auditor

Goal

Detect semantic mismatches that pass schema checks but break assessment validity or grading fairness.

Inputs

mission.json
statement.md
rubric.md
teacher-notes.md
Optional reference/ and tests/

Produce

Write semantic-consistency-report.json with:

overall_status (pass|review|fail)
findings (id, severity, artifact, evidence, remediation)
coverage_matrix (learning outcomes x rubric criteria x tasks/tests)
leakage_risks (teaching/exam mismatch, hidden prerequisites, rubric blind spots)
fairness_risks (ambiguous wording, non-deterministic grading, time-budget mismatch)

Mandatory checks

Every assessed task maps to at least one declared learning outcome.
Every rubric criterion maps to observable evidence in statement or expected outputs.
Teacher notes do not introduce criteria absent from rubric.
Tests (if present) verify intended competencies and do not overfit incidental details.
Time/difficulty estimate is consistent with mission constraints.

Severity policy

fail: invalidates grading integrity (missing alignment, contradictory criteria, impossible constraints).
review: ambiguous but potentially salvageable.
pass: no blocking inconsistency detected.

Rules

Prioritize evidence-based findings over style feedback.
Report concrete remediation steps, not generic advice.
Never mutate source artifacts; emit audit only.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.