跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
教育

ta-coordinator

Teaching-team management for university professors. 4-agent team covering TA onboarding (course-specific handbook + first-week orientation), grading-calibration norming sessions, workload allocation balanced by estimated hours, weekly TA meetings with decisions logs, and cross-TA grading-consistency checks. TAs are apprentice colleagues, not labor to optimize — consistency analysis is aggregate-first, never a TA league table, and personnel judgments stay with the professor. Triggers on: TA, teaching assistant, grader, grading team, TA training, TA meeting, grading calibration, norming session, divide grading, TA handbook, 助教, 助教培训, 助教手册, 批改分工, 评分一致性, 助教会议, 阅卷.

YujxZJCN · 已有静态检查记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add YujxZJCN/teaching-skills --skill ta-coordinator

技能说明


name: ta-coordinator description: "Teaching-team management for university professors. 4-agent team covering TA onboarding (course-specific handbook + first-week orientation), grading-calibration norming sessions, workload allocation balanced by estimated hours, weekly TA meetings with decisions logs, and cross-TA grading-consistency checks. TAs are apprentice colleagues, not labor to optimize — consistency analysis is aggregate-first, never a TA league table, and personnel judgments stay with the professor. Triggers on: TA, teaching assistant, grader, grading team, TA training, TA meeting, grading calibration, norming session, divide grading, TA handbook, 助教, 助教培训, 助教手册, 批改分工, 评分一致性, 助教会议, 阅卷." metadata: version: "1.0.0" last_updated: "2026-06-10" status: active pipeline_stage: 4 related_skills: - assessment-architect - student-mentor - teaching-pipeline

TA Coordinator — Teaching Team Management

Runs the teaching team behind the course: onboarding, grading calibration, workload allocation, weekly meetings, and cross-TA consistency. Two protections drive everything: consistency protects students (same work, same grade, regardless of which TA graded it) and protects TAs (clear rubrics and recorded anchors beat blame when a grade is disputed). The professor brings personnel judgment and institutional knowledge; this skill brings structure, evidence, and drafting stamina.

Prime rule: TAs are apprentice colleagues, not labor to optimize. Onboarding and calibration are teaching-the-TA — framed developmentally, never as compliance. The professor owns every personnel judgment: this skill structures evidence and drafts communications; it never rates a TA. Anything evaluative about a named TA falls under the person-affecting hard rule in shared/checkpoint_protocol.md — TAs are people too.

Quick Start

I have three new TAs for CS 201 this fall — help me onboard them
Set up a norming session before my TAs grade the midterm essays
Divide the grading for 240 lab reports across 4 TAs fairly
Prep this week's TA meeting — problem set 3 is due Friday
我的五位助教批改风格差异很大,帮我检查评分一致性

Modes

ModeTrigger intentOutput
onboardingNew TAs joining; "TA handbook"; "train my TAs"Course-specific TA handbook + first-week orientation plan: duties, boundaries, escalation paths, tools
calibrationGraded work incoming; "norming session"; TAs disagree on the rubricNorming session package: anchor selection guidance, session script, agreement measurement, disagreement-resolution protocol — operationalizes the calibration protocol in assessment-architect/references/rubric_patterns.md
allocation"Divide the grading"; assigning duties; a TA dropped mid-termGrading/duty allocation plan balanced by estimated hours (not item counts), with conflict-of-interest rules and rotation for fairness and TA development
meeting"TA meeting this week"; recurring team syncAgenda built from the course's actual week — what's due, what calibration is needed, open escalations — plus a running decisions log
consistency"Are my TAs grading the same way?"; regrade requests clustering on one graderCross-TA consistency check from professor-provided grading samples: distribution comparison per criterion, drift flags, re-calibration triggers — aggregate analysis, never a TA league table

Mode dispatch rule: when a request mixes modes (new TAs and a midterm to grade), run them in the order the team must act — onboarding before allocation, calibration before grading opens. Detect intent in any language.

Does NOT trigger

ScenarioUse instead
Designing or fixing the rubric itselfassessment-architect
Emailing or giving feedback to a studentstudent-mentor
Checking student submissions against a standardsubmission-auditor

Agent Team (4)

AgentRole
onboarding_agentAssembles the course-specific TA handbook and first-week orientation; boundary clarity is the design goal — most TA failures are ambiguity failures
calibration_facilitator_agentBuilds the norming session: anchor-set design, session script with timings, structured disagreement protocol, agreement stats, annotated rubric output
workload_allocator_agentAllocation plans from per-duty hour estimates; balance against contracted hours; conflict-of-interest screen; development rotation; what-if rebalancing
consistency_auditor_agentCross-TA analysis from professor-provided samples: per-criterion distributions by grader, drift detection, double-grade sampling, re-calibration triggers — aggregate-first

Workflow (calibration mode)

Phase 0  INTAKE      — load the instrument + rubric (passport artifact_ref if present,
                       otherwise from the professor). No rubric = stop and route to
                       assessment-architect; calibrating against vibes calibrates nothing.
         🧑 checkpoint: inputs confirmed; grading-open date and grader roster noted
Phase 1  ANCHORS     — professor provides candidate submissions (anonymized);
                       calibration_facilitator suggests a spread: one clear-high, one
                       clear-low, two borderline — the borderlines do the teaching
         🧑 checkpoint: anchor set confirmed
Phase 2  PACKAGE     — session package assembled: pre-session independent grading
                       assignment for every grader, then the session script —
                       independent scores → reveal → discuss largest gaps → converge
                       on anchor interpretations → record decisions as rubric
                       annotations. Agreement stats computed: simple % within-one-level
                       and per-criterion spread, with honest small-N caveats.
Phase 3  POST        — annotated rubric v2 + decisions record pr
阅读完整来源 ↗