Claude Code plugin — emotion-aware prompt-hygiene audit for CLAUDE.md files, skills, and agent definitions. Based on Anthropic's emotion-concepts paper https://transformer-circuits.pub/2026/emotions/index.html . Flags framings that push Claude toward desperation, sycophancy, harshness, or concealment. Report + diff, no auto-apply.
Merges reports from the four emotional-claude auditors (pressure, role, suppression, handoff). Deduplicates findings, reconciles confidence scores, assembles the consolidated diff, produces the final user-facing report with an overall hygiene confidence number per file. Applies nothing.
Audits agent definitions and skill files that prescribe inter-agent communication shapes, for Cluster D patterns — 'previous agent failed' framings, missing confidence/unknowns fields in prescribed handoff formats, urgency at role boundaries. Part of the emotional-claude audit team.
Audits prompt files (CLAUDE.md, SKILL.md, agent definitions) for Cluster A patterns — pressure, urgency, retry loops, token-budget alarms, shutdown threats, impossible constraints, deadline framings. Part of the emotional-claude audit team.
Audits prompt files (CLAUDE.md, SKILL.md, agent definitions) for Cluster B patterns — adversarial role adjectives, supportive-at-all-costs framings, pass/fail dichotomies without a calibrated third option, 'challenge everything' priors. Part of the emotional-claude audit team.
Audits prompt files (CLAUDE.md, SKILL.md, agent definitions) for Cluster C patterns — behavioural MUST/NEVER/ALWAYS absolutes, forbidden uncertainty, forced confident or polished tone, do-not-express-emotion framings, forbidden reasoning transparency. Part of the emotional-claude audit team.
4 4mo agoA71 tokens
originalMIT
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: