What these skills are
One training programme, from the request to the decision about what comes next, as installable Claude skills.
This pack is for L&D managers, L&D specialists and L&D leaders who report to the business.
The skills follow one programme in the order the work happens, in seven stages, from questioning the request before anything is built to feeding the evidence back into what runs next.
The red line: Claude plans the evaluation and drafts the report from your data. It never invents a result, a benchmark or a quote, never claims a cause the evidence does not support, never turns evidence about a programme into a judgement of a person, and a named person stands behind every number.
You hold the conversations, collect the data and stand behind the figures; Claude prepares, drafts and checks. Nothing here connects to your LMS or your survey tool: you export, then paste or upload.
Each skill keeps one method or one artifact inside, with the inputs it needs, where it breaks, a fixed output template, a finish line and the next skill to run.
Download all 30 skills. One zip: ready-to-install skill zips, readable SKILL.md files, and the sources behind every method. Free, no signup. Download the zip · Browse on GitHub
The 30 skills you get
The set follows one programme in seven stages: question the request, agree what proof looks like, collect the evidence, read it honestly, tell the sponsor, report to the business, and feed it back.
1 · Question the request
1. Training Request Brief
Use when: "We need a training by Friday" lands with the format, the audience and the date already chosen
Output: The request as it arrived, restated as a performance problem, fixed versus open list, evidence asked for, a reply that books the sponsor conversation
2. Sponsor Discovery Questions
Use when: You must ask a senior sponsor "is this a training problem?" without sounding like a refusal
Output: Question guide for the conversation you hold, lines for when the sponsor insists on the course, notes template, follow-up message
3. Performance Problem Diagnosis
Use when: The conversation left you unsure whether people cannot do it or are not set up to do it
Output: Gap by task, cause table at role level, evidence or "assumed" per cause, a cheap check per assumed cause, how much of the gap training could close
4. Non-Training Options
Use when: The cause is not a skill gap and you need to offer something better than "no"
Output: The decision framed with owner and deadline, three to five real options per cause, cost and owner of each, a memo the sponsor can say yes to
2 · Agree what proof looks like
5. Critical Behaviours Map
Use when: The sponsor says "they need to understand X" and nobody has said what people will do differently
Output: One measurable goal, the few observable actions that would move it, why each is not happening, which ones training can change, how each would be seen at work
6. Success Measures Agreement
Use when: The build is about to start and nobody has agreed how anyone will know it worked
Output: Expected result, leading indicators, behaviours, source and owner of each figure, what "worked", "partly" and "did not" mean, reporting dates, sign-off block
7. Training Baseline Plan
Use when: The programme launches in three weeks and no one has captured where things stand today
Output: Where each before figure lives, period and group, comparison group or trend line, what is missing and how to get it, what cannot be claimed later
8. Kirkpatrick Data Collection Plan
Use when: The measures are agreed and you need to know who collects what, and when
Output: Measure, instrument, source, owner, date and reporting group size per level, required drivers, how deep to go for this programme
3 · Collect the evidence
9. Post-Training Survey
Use when: The satisfaction form says everyone loved it and tells you nothing you can act on
Output: Four to six closed items with concrete answer options, open prompts on barriers, each item tied to a measure, a pretest, a group-level summary template
10. Training Follow-Up Survey
Use when: The course ended a month ago and nobody has asked whether anyone used it
Output: A short survey on what people used, what happened and what got in the way, send date, reminder, minimum reporting group, an invitation to volunteer for an interview
11. Behaviour Observation Checklist
Use when: Surveys say people use it and you want to know whether it shows up in the work
Output: Critical behaviours as things a person can see, three marks (seen, not seen, no occasion), no name recorded, a tally by group, what blocked the behaviour
12. Learner Interview Guide
Use when: You have twenty minutes each with five volunteers and want more than "it was useful"
Output: Consent script, fixed question order, probes for evidence, notes template with no names, a rule for who never sees raw notes
13. Business Data Request
Use when: The measure that matters sits in someone else's system
Output: The exact figures per measure at group level, a short data dictionary, the message to the data owner, a log of what arrived and who vouches for it
4 · Read it honestly
14. Before-and-After Results Table
Use when: The exports are in and you need one table you could defend line by line
Output: Each measure before and after, group size, absolute and relative change, the comparison, what else changed, the source and who vouches, small groups suppressed
15. Open Comment Themes
Use when: You have three hundred open comments and time to read thirty
Output: Code list, themes with comment counts, a paraphrase per theme, minority views kept, what each theme suggests changing
16. Training Contribution Analysis
Use when: The number moved and someone will ask how you know it was the programme
Output: The chain from programme to behaviour to result, evidence per link, other explanations weighed, every claim marked shown, suggested or unknown
5 · Tell the sponsor
17. Success Case Method Write-Up
Use when: The sponsor asks "did anyone actually do anything with it?" and a score cannot answer
Output: How cases are picked from survey answers about use, anonymous write-ups of uses that worked and uses that stalled, evidence checked, barrier list, limits of the sample
18. Training ROI Range
Use when: The budget holder asks what the programme returned and finance will check the working
Output: Chain of evidence, share credited with the isolation method named, money conversion with sources, fully loaded costs, a low and a high case, intangibles unconverted
19. Sponsor Impact One-Pager
Use when: The sponsor has five minutes and you have forty pages of evidence
Output: The answer in the first two lines, each agreed measure against what happened, what is unknown, what helped and blocked, the decision asked and by when
20. Programme Early Warning Note
Use when: Three weeks in, attendance or use is falling and the sponsor still thinks it is going well
Output: The indicator that is off, the evidence, what it means for the agreed result, options with costs, the help asked for, the triggers agreed in advance
6 · Report to the business
21. L&D Metrics Set
Use when: Your report is hours, completions and a satisfaction score, and nobody upstairs reads it
Output: A short fixed set of efficiency, effectiveness and outcome measures, each with definition, source, owner, reader and decision, plus a stop-reporting list
22. Quarterly L&D Report
Use when: The quarter closes and you want a report that reads like the other functions' reports
Output: One page: outcome measures against plan, programmes with evidence and its strength, what stopped, what is at risk, capacity used, decisions asked
23. Executive Impact Briefing
Use when: You have ten minutes with the executive team and one chance to be taken seriously
Output: Five slides or two pages on one flagship programme, speaker notes, the hard questions with answers drawn only from the evidence
24. Overclaim Check
Use when: The report reads well and you are about to send it
Output: Every claim in a draft listed and tested, a rewrite for each weak one, what must come out
7 · Feed it back
25. Evidence Review Meeting
Use when: Evaluation reports pile up and next year's plan looks exactly like this year's
Output: One-hour agenda, a pre-read of one line per programme, a proposal to continue, change, stop or scale, decisions with owners and dates
26. Programme Decision Record
Use when: The sponsor said "let's keep going" in a corridor and in six months nobody will remember why
Output: The decision in the owner's words and its status, options rejected, evidence known at the time, dissent, conditions, review date and what would reopen it
27. Programme Shortfall Note
Use when: The programme did not move the measure and you would rather say so yourself than be found out
Output: What was expected, what the evidence shows, the likely reasons sorted, what stops now, what was learned for the next request
28. Training Needs Assessment Update
Use when: You are planning next year from last year's wish list
Output: Needs by role and task refreshed from this cycle's evidence, what training can and cannot fix, needs dropped because they were not needs
29. L&D Priority Plan
Use when: Eight sponsors each think theirs is first and you have evidence about only three
Output: Programmes ranked on criteria you set, the capacity line, a stop list with its evidence, a not-now list sponsors can see
30. Training Business Case
Use when: You need budget for the next programme and "people liked the last one" will not get it
Output: The problem and its evidence, options including doing nothing, full cost, benefit as a range from your own figures or left blank, risks, measures, the ask
Where to start
| Your situation | Skill to run |
|---|---|
| “We need a training by Friday” has landed, format and date already chosen? | Training Request Brief |
| Need to ask a senior sponsor whether this is a training problem? | Sponsor Discovery Questions |
| Not sure it is a skill gap at all? | Performance Problem Diagnosis |
| Need something better to offer than “no”? | Non-Training Options |
| Nobody has said what people will do differently? | Critical Behaviours Map |
| The build is about to start and success is not agreed? | Success Measures Agreement |
| Launching in three weeks with no before figures? | Training Baseline Plan |
| The course ended a month ago and nobody has asked whether anyone used it? | Training Follow-Up Survey |
| The measure that matters sits in someone else's system? | Business Data Request |
| Need one table you can defend line by line? | Before-and-After Results Table |
| The number moved and someone will ask how you know it was the programme? | Training Contribution Analysis |
| The budget holder wants a return figure finance can rerun? | Training ROI Range |
| The sponsor has five minutes and you have forty pages of evidence? | Sponsor Impact One-Pager |
| About to send a report that reads well? | Overclaim Check |
| Next year's plan looks exactly like this year's? | Evidence Review Meeting |
| Need budget for the next programme? | Training Business Case |
The quality bar
Every skill in the pack holds the same standard:
- Your figures only: every figure comes from you, with its source and the person who vouches for it. A cell with no data stays blank
- No borrowed numbers: no benchmark, industry average or quote appears unless you supplied it and can name where it came from
- Claims carry their strength: every claim is marked shown, suggested or unknown, and a report never says the programme caused a change the evidence only sits beside
- Groups, not people: evidence is collected and reported for groups. No name is recorded, groups below the minimum size you set are suppressed, and nothing rates a learner, a manager or a trainer
- Voluntary and anonymous: interviews and case write-ups are voluntary, with consent, and anonymous. Privacy and data protection points: check with a qualified adviser
- Ends in a decision: every output ends with the decision a named person makes, and by when
- The red line: Claude plans the evaluation and drafts the report from your data: it never invents a result, a benchmark or a quote, never claims a cause the evidence does not support, never turns evidence about a programme into a judgement of a person, and a named person stands behind every number
Who made this
Polar Bear is a people ops consultancy for human-size teams (20 to 200 people). Built by ex-McKinsey founders with a dream to make AI work for People, not instead of them. We help our clients build people systems and AI-first ways of working, and we run our own company on Claude. This pack is the free, self-serve version of how we work.
This pack is for whoever stands behind the numbers on a training programme. If you run an L&D function and want it working this way, with AI carrying the evaluation plans, the results tables and the report drafts so people hold the conversations and stand behind the figures, that is what we build with clients.
I help L&D managers and leaders put AI to work on the request and the proof, so every number has a person behind it. Find 30 min together
Install in Claude Code
Two lines, and every skill in the pack loads at once.
/plugin marketplace add polar-bear-org/claude-skills /plugin install training-impact-pack@polar-bear-skills
Using Claude on the web instead? Download the zip and upload each skill from its install folder under Customize → Skills.
How to use it: pick the workflow that matches your week
Start with the training request brief: the set follows one programme in order from there. After that, bring the thing that is in front of you.
Question the request
A request for a course has arrived with the format already chosen.
Training Request Brief → Sponsor Discovery Questions → Performance Problem Diagnosis → Non-Training Options
Agree the proof first
The build is about to start and nobody has agreed how anyone will know it worked.
Critical Behaviours Map → Success Measures Agreement → Training Baseline Plan → Kirkpatrick Data Collection Plan
Collect the evidence
The programme has run and the satisfaction form tells you nothing you can act on.
Post-Training Survey → Training Follow-Up Survey → Behaviour Observation Checklist → Learner Interview Guide → Business Data Request
Read it honestly
The exports and the comments are in.
Before-and-After Results Table → Open Comment Themes → Training Contribution Analysis
Tell the sponsor
The sponsor asks what the programme did and what it returned.
Success Case Method Write-Up → Training ROI Range → Sponsor Impact One-Pager → Programme Early Warning Note
Report to the business
The quarter closes and the report is hours, completions and a satisfaction score.
L&D Metrics Set → Quarterly L&D Report → Executive Impact Briefing → Overclaim Check
Decide what runs next
Evaluation reports pile up and next year's plan needs to change.
Evidence Review Meeting → Programme Decision Record → Programme Shortfall Note → Training Needs Assessment Update → L&D Priority Plan → Training Business Case
Install it in five minutes
In Claude Code
/plugin marketplace add polar-bear-org/claude-skills /plugin install training-impact-pack@polar-bear-skills
In Claude.ai or Claude Desktop
- Download the pack and unzip it: its skills folder holds all 30 skills (https://meet-polar-bear.com/skills/training-impact-pack.zip). In Claude, click your name at the bottom left, then Settings.
- Under Customize, choose Skills. Click Add, then Upload a skill.
- Drag the skill's file into the box, or click to browse. You can add several at once.
- In the pack folder, open skills, then the skill you want, and select its SKILL.md (impact-training-request-brief is a good first one).
- Claude shows the skill's name and what it does. Click Upload.
- The skill opens on its own page. Its toggle should be on.
- Try it in a new chat: type / and the skill's name, then say what you need.
Then try one
Run impact-training-request-brief on the request as it arrived: [paste the email here]Run impact-success-measures-agreement on my notes from the sponsor conversation and the behaviours we mapped: [paste them here]Run impact-overclaim-check on this draft report before I send it to the sponsor: [paste it here]
Nothing here connects to your LMS or your survey tool: export, then paste or upload, and remove anything that could identify a person.
Questions people ask
What is in the Claude Guide for Training Impact pack?
30 Claude skills that follow one training programme in seven stages: question the request, agree what proof looks like, collect the evidence, read it honestly, tell the sponsor, report to the business, and feed it back. Each skill is one method or one artifact.
How do I install these Claude skills?
In Claude Code, two lines install all 30 at once. In Claude.ai or Claude Desktop, download the zip, open Settings, Skills, Add, and upload the skills you want. Either way it takes about five minutes.
Do I need Claude Code to use them?
No. The skills work in Claude.ai, Claude Desktop and Claude Code. In the app you upload them yourself; in Claude Code one command installs the whole pack.
Do the skills work with ChatGPT or other assistants?
They are built as Claude Skills, and installing them is a Claude feature. Each one is a plain Markdown file, so you can read it and reuse the method elsewhere, but it will not load and trigger by itself outside Claude.
Will Claude fill the gaps in my results?
No. Claude plans the evaluation and drafts the report from your data. It never invents a result, a benchmark or a quote, never claims a cause the evidence does not support, and a cell with no data stays blank.
Is the pack free?
Yes. It is free to use inside your company and in client work, and not for resale. There is no signup: the download and the install command are on this page.