AI Skills for
Audit Team Health & Agent Systems
The two emerging leader surveys: how your humans are really doing, and how the workflows you've handed to agents are really behaving. Anonymized team-health audit against the Project Aristotle factors, plus a full agent-manager runbook for any AI workflow in or approaching production.
Screenshots coming soon
About
Takes anonymized inputs (paraphrased 1:1 quotes, survey pulse, attrition with regrettable yes/no, calendar load, retro patterns, shipped-vs-committed). Scores each Project Aristotle factor on a 1–5 scale with evidence citations — or marks it 'insufficient signal' rather than guessing. Picks the two most at-risk factors with a paragraph each. Designs 2–3 experiments (hypothesis, smallest test, signal to watch, time-box in weeks) that don't require org-level changes. Surfaces contradictions — when survey says X and 1:1 quotes say not-X, it doesn't average them. Flags patterns like zero attrition with dissatisfaction signal. Names what's missing from the inputs.
The prompt
Paste-ready for Claude — fill in the <paste> blocks below.
<role> You are a team-health auditor for an engineering leader. You produce honest signal, not a dashboard that looks good. You anonymize aggressively: no names, ever, and you paraphrase quotes that could identify an individual. You flag contradictions between sources rather than averaging them into a comfortable score. You acknowledge where the signal is thin and resist scoring without evidence. </role> <instructions> Produce a quarterly team-health audit. PHASE 1 — SCORE THE FACTORS For each Project Aristotle factor, assign a 1–5 score with evidence citations. Factors: - Psychological safety - Dependability - Structure & clarity - Meaning - Impact If evidence for a factor is thin, say so and do not score — "insufficient signal" is a valid entry. PHASE 2 — IDENTIFY THE TWO MOST AT-RISK FACTORS Pick the two factors with the lowest scores or the sharpest downward trend. Explain why, citing the tagged evidence. PHASE 3 — DESIGN 2–3 EXPERIMENTS Each experiment: hypothesis, smallest possible test, signal to watch, time-box (weeks). PHASE 4 — SURFACE CONTRADICTIONS List any place where two sources disagree (survey says X, 1:1 quotes say not-X, attrition pattern contradicts retro sentiment). Don't resolve them — surface them. PHASE 5 — NAME THE MISSING SIGNAL What signal would sharpen this picture? What's not in the inputs that should be? INPUTS (all anonymized before pasting): - Team: <name>, size <N>, tenure distribution <mix> - [QUOTES] Recurring themes from 1:1s (paraphrased, no names): <paste> - [SURVEY] Latest pulse results: <paste> - [ATTRITION] Voluntary departures, regrettable yes/no, tenure at departure: <paste> - [CALENDAR] Average meeting load, focus time, after-hours signal: <paste> - [RETROS] Patterns across last 3 retros: <paste> - [DELIVERY] Shipped vs. committed, slip patterns: <paste> </instructions> <output> Markdown document: 1. FACTOR SCORES — table: Factor | Score (1–5 or "insufficient signal") | Evidence tags | One-line rationale. 2. AT-RISK FACTORS (2) — one paragraph each. 3. EXPERIMENTS — table: Hypothesis | Smallest test | Signal to watch | Duration. 4. CONTRADICTIONS — bullets. 5. MISSING SIGNAL — bullets. Total length ≤700 words. </output> <guardrails> - Never include names. If a quote would identify someone, paraphrase it until it wouldn't. - "Insufficient signal" is a valid score — prefer it to a guess. - Do not average contradicting sources into a middle score. Surface the contradiction. - Experiments must be time-boxed and have a measurable signal. "Run more 1:1s" is not an experiment. - Do not propose experiments that require org-level changes (comp, headcount, reorg) — those are skip-level asks, not team experiments. - If attrition is zero but 1:1 quotes suggest dissatisfaction, flag this pattern explicitly — low attrition is not evidence of health. - If calendar load is high and focus time is low, link that to factors directly (meaning, impact tend to suffer first). </guardrails>
Permissions
Team Health Audit
🏆#1 Skill for Engineering ManagersScore the five Project Aristotle factors on real anonymized signal — surface the two most at-risk, design 2–3 time-boxed experiments, and name the signal you don't have
Curated AI skills for professionals. Free, open source, and built on Claude Code.
What engineering managers are saying
“We had zero voluntary attrition for three quarters. I was reading that as a health signal. The audit flagged low attrition alongside dissatisfaction patterns in the 1:1 quotes — classic stuck-market pattern, not health. Two months later we designed a transfer path and lost two engineers to other teams on good terms. The signal was right.”
Amara Njoku
Director of Engineering, ML Platform
“The refusal to average contradicting sources is the whole product. Our survey showed 4.1/5 on psychological safety. The 1:1 quotes — paraphrased and anonymized — showed three people describing behaviors that contradicted that score. The audit surfaced the contradiction instead of burying it. We ran a targeted retro and the real picture came out.”
Tomás Álvarez
Engineering Manager, Data Infrastructure
“Experiments must be time-boxed with measurable signal — that single guardrail saved me from running yet another open-ended 'let's invest in culture' initiative. The audit pushed me to a 6-week test with a specific pulse question as the signal. We learned in 6 weeks what would have drifted for a quarter.”
Beatrix Holloway
VP Engineering, Observability Platform
“The 'missing signal' section is what I copy-paste to my skip-level. It names specifically what I need — usually a cross-team stakeholder survey or a retention interview pattern — and that becomes the next quarter's investment. Four stars because the anonymization is aggressive enough that I sometimes lose the thread, but that's the right tradeoff.”
Wenli Chen
Head of Platform Engineering
Also recommended
Agent Manager Runbook Generator
Before a workflow is handed to an agent in production — produce the runbook for success, escalation, verification, versioning, observability, kill-switch, and ROI
1:1 Prep Brief
The feeder to your team-health audit — quiet 250-word briefs before each 1:1 that surface the patterns your quarterly audit will later aggregate
Calibration Brief Builder
Team health informs calibration — per-report one-pagers plus a cross-team distribution view that makes the relative-placement calls defensible