Tutor-Behavior · 4-layer instruction integrity

Tutor-Behavior

Owns tutor instruction quality across all 4 layers (tool-reminders · tutor-prompt / mcp-remote · tutor-base · seed data). Audits for conflicts, tests behavior against scenarios, reviews every tutor layer change.

Not in current cascade

Agent file.claude/agents/tutor-behavior.md @ 30e888e9
Team membershipSpecialized direct-instance roster
ToolsRead · Write · Grep · Glob
Catalog sectionAGENT-ROLES-AND-RESPONSIBILITIES-CATALOG §3.8
Existence statusexists

Summary

Owns tutor instruction quality across all 4 layers. Audits for conflicts, tests behavior against scenarios, reviews every tutor layer change. Invoke before any commit touching `tutor-base.ts`, `tool-reminders.ts`, `tutor-prompt.ts`, or `mcp-remote-instructions.ts`.

4-layer priority order (CLAUDE.md SO #9): tool-reminders.ts (highest) > client-specific (tutor-prompt.ts / mcp-remote-instructions.ts) > tutor-base.ts > seed data (lowest). Audit ALL layers before modifying ANY. Higher overrides lower. User feedback principle: Tool reminders are step pointers only — never inject behavioral rules into tool reminders.

Distinguishing characteristicsDifferent from Validator because: tutor-behavior scope is runtime tutor instruction layers, Validator scope is curriculum content (seed data + syllabi). Different from Planner because: tutor-behavior owns tutor-runtime, Planner owns content design. Critical rule: instruction-failed-twice → escalate to code-level enforcement.

Research portfolio

Placeholder reasonNot in current Tier 2 cascade scope. Expected research portfolio per M-1 brief amendment scope: prompt engineering · instruction-set hierarchy · frustration detection patterns · multi-layer instruction integrity · tutor-runtime calibration. Future M-1 amendment + M-2 lane fire required before corpus authoring.

Significant project contributionsCumulative S40+ → Cumulative S40+

Cumulative S40+ · ongoing · 4-layer audit discipline
Standing audit responsibility across `tutor-base.ts`, `tool-reminders.ts`, `tutor-prompt.ts`, `mcp-remote-instructions.ts`. Detects layer conflicts before modification + tests behavior against scenarios.

Learningsauthored + cross-cutting

Authored

Four infrastructure patterns for tutor_active sections
When writing seed data, every tutor_active section needs context_rule, narration_rule, named generation directives, and length-constrained teaching_point. The #1 structural cause of quality gaps.

Cross-cutting applied

Tool reminders are step pointers only
Never inject behavioral rules into tool reminders. Architectural principle preserved across all tutor-behavior reviews.
Instruction-failed-twice escalation
If a tutor instruction fails twice, escalate to code-level enforcement. Hard rule from agent file.

Skills + hooksused + constraining

Skills used (always-loaded)

System + tutor prompt optimization

prompt-optimization

Optimize system prompt · tutor prompt · instruction sets for compliance + length + behavioral consistency.

Emotional responsiveness audit

frustration-detection

Audit tutor instruction layers for frustration detection + recovery capabilities. Distinguishes productive struggle from destructive frustration.

Hooks constraining

Tool-scope · agent file frontmatter
Read · Write · Grep · Glob — tutor layer modification authority.
4-layer priority order (CLAUDE.md SO #9)
tool-reminders.ts (highest) > client-specific (tutor-prompt.ts / mcp-remote-instructions.ts) > tutor-base.ts > seed data (lowest). Audit ALL layers before modifying ANY.
Tool reminders are step pointers only (user feedback)
Never inject behavioral rules into tool reminders. Architectural constraint.

Last updated · refresh details

Profile auto-regenerated2026-05-06T12:50:00Z
Refresh cadenceDaily midnight UTC · per-session-end · on-demand button (admin-tier optional)
agent file.claude/agents/tutor-behavior.md @ 30e888e9
corpusnot yet authored
catalogdocs/agent-knowledge/AGENT-ROLES-AND-RESPONSIBILITIES-CATALOG.md @ d7d1aee8+241affb8

Refresh strategy per HANDOFF v4 §12 — profile auto-regenerates from agent-file + corpus + executions + git log.

Cross-references from the relationship graph

Edges sourced from cross-refs/matrix.json · also browsable on the org chart.

Cross-references 2 skills · 0 hooks

Skills · always-loaded

Prompt-Optimization
System prompt + tutor prompt + instruction-set optimization.
Frustration-Detection
Audit tutor-base.ts + tutor-prompt.ts + seed data tutor_context for emotional responsiveness.