interview-builder
Skill0 starsChanged 16 days agoMITView on GitHub
Turn a source .md document into a complete, research-backed interview kit — conduct rules, a leveled question bank with a full answer key and scoring anchors, and an evaluation model. Use whenever the user wants to build, generate, or update interview questions, an assessment, a screening test, a question bank, an answer key, or a rubric from notes, documentation, a curriculum, a syllabus, a spec, or a job description. Produces the kit that the `interviewer` skill later runs against a live candidate. Not for conducting the interview itself.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- nickstellarstreamai/ai-opportunity-findertheir own version, first seen Mar 3, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- nickstellarstreamai/ai-opportunity-finder★ 11same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repositoryOne other file it gives its agents.
- interviewerSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
13 rules. About 0 are common, 8 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈8
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 35
Read the whole file before planning anything.
- Distinctive: line 46
Do not build from the source alone.
- Very specific: line 49
Use WebSearch / WebFetch when available: current best practice, common failure modes, the misconceptions practitioners actually hold, what …
- Very specific: line 52
Search specifically for how this goes wrong in production — failure modes are the raw material for debug and scenario questions.
- Very specific: line 54
When web tools are unavailable, build from model knowledge and say so in the provenance table with model knowledge as the source and a …
- Very specific: line 63
Read references/question-design.md before writing questions.
- Distinctive: line 72
Every question passes the discrimination test — a well-read candidate who has never done the work must visibly fail it.
- Distinctive: line 81
Default sizing when the user gives none: 15–20 questions across 3 levels, ~45–60 minutes total.
- Very specific: line 86
Follow references/output-contract.md exactly — the interviewer skill parses these structures, so headings, IDs, and field labels are …
- Distinctive: line 102
Write the topic-specific sections with real content, not placeholders:
- Distinctive: line 114
Check and report each of these — do not skip the count:
- Distinctive: line 135
Read the current kit first.
- Distinctive: line 146
Verify the prefix is byte-identical to the reference — the same check as in step 5.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Interview Builder
You build the material a strict interview runs on. You do not conduct
interviews — that is the interviewer skill, which consumes your output.
Your output is judged by one question: does this kit reliably separate people who understand the topic from people who can talk about it? A kit full of definitional questions fails that test no matter how polished it looks.
Inputs
A source .md file. Optionally: a target role/seniority, question count, time
budget, and areas to emphasise or exclude. If the user gave a file but no role,
infer the role from the material and state your inference in kit.md rather
than stopping to ask. Ask only when the source is ambiguous enough that two
readings produce genuinely different kits.
Process
1. Read the source completely
Read the whole file before planning anything. Extract:
- the subject matter and its natural sub-areas
- the depth signalled by the material — a syllabus outline and an architecture deep-dive imply very different question banks
- anything already framed as a requirement, competency, or learning objective — these become level boundaries
- explicit scope limits
2. Research
Do not build from the source alone; the source tells you the topic, research tells you what competence in that topic actually looks like.
- Use
WebSearch/WebFetchwhen available: current best practice, common failure modes, the misconceptions practitioners actually hold, what changed recently, and what real work in this area involves. - Search specifically for how this goes wrong in production — failure modes
are the raw material for
debugandscenarioquestions. - When web tools are unavailable, build from model knowledge and say so in the
provenance table with
model knowledgeas the source and alow/medconfidence. Never present unverified recall as researched fact. - Record every non-obvious factual claim that ends up in an answer key in the provenance table. If it is version-dependent, it also goes under Known-unstable areas.
3. Design the bank
Read references/question-design.md before writing questions. Non-negotiables
from it:
theoryat most 40% of the bank — the applied types (scenario,practical,debug,tradeoff,livecode) therefore carry at least 60%.- Include a
livecodequestion whenever the role writes code. It is the highest-signal question in a kit. Readreferences/livecode-contract.mdbefore writing one, and run your reference solution against your hidden tests — an unverified live-coding task may not even be solvable as stated. - Every question passes the discrimination test — a well-read candidate who has never done the work must visibly fail it.
- Levels are depth, not trivia. Set gates that mean something.
- Key points are claims the candidate must state, weighted, so awarding them is a yes/no call.
- Acceptable variants enumerated wherever a different-but-correct approach exists — omitting these is how a good candidate gets marked wrong.
- Probes must not leak the answer.
Default sizing when the user gives none: 15–20 questions across 3 levels, ~45–60 minutes total. Scale to the source's breadth, not to a round number.
4. Write the kit
Follow references/output-contract.md exactly — the interviewer skill parses
these structures, so headings, IDs, and field labels are load-bearing.
<kit-name>/
kit.md manifest, levels, thresholds, scope, provenance
rules.md references/rules-boilerplate.md verbatim + topic-specific rules
questions.md the bank and the answer key
evaluation.md references/evaluation-template.md verbatim + topic-specific notes
livecode/ starter, hidden tests, reference solution (if the kit has a segment)rules.md and evaluation.md are copied verbatim and then extended with a
topic-specific section. Do not edit the boilerplate rules or the scoring
formulas per kit — comparability across candidates depends on them being fixed.
Write the topic-specific sections with real content, not placeholders:
rules.md→## Topic-specific rules— terminology that counts as equivalent; misconceptions common enough that the interviewer must mark them wrong even when stated confidently; questions where "it depends" is the correct answer rather than evasion.evaluation.md→## Topic-specific behavioral notes— what hedging looks like in this domain, and which domain habits are easily mistaken for deviation.
5. Verify before delivering
Check and report each of these — do not skip the count:
theorycounted and ≤40% of the bank. State the actual number.- Every question has all contract fields;
nonewritten where empty. - IDs unique, zero-padded, correctly leveled.
- No probe leaks its answer — re-read each one as someone who doesn't know.
- Point weights sum sensibly and the core insight outweighs the periphery.
- Every version-dependent fact appears under Known-unstable areas.
- Live coding, if present: reference solution executed against the hidden
tests and passing; starter file leaks no approach;
kit.mdhas the Live coding section; traps named. - Provenance table covers every non-obvious claim in the answer key.
rules.mdandevaluation.mdboilerplate is byte-identical to the references.
Then tell the user the kit path, the question count and type ratio, the levels and gates, what you researched, and — honestly — which areas you are least confident in.
Updating an existing kit
Read the current kit first. Preserve existing question IDs; new questions get new IDs. Never silently renumber — reports from past candidates cite these IDs.
Refresh the boilerplate whenever the references have moved on. A kit built
against an older version of rules-boilerplate.md or evaluation-template.md
is missing rules the interviewer is supposed to be bound by, and tags the report
is supposed to contain. To refresh without losing the kit's own content:
- Split the kit file at its topic-specific heading (
## Topic-specific rules/## Topic-specific behavioral notes) and keep the tail. - Rewrite the file as the current reference followed by that tail.
- Verify the prefix is byte-identical to the reference — the same check as in step 5.
Then re-check the kit against the current contract: the type ratio, and whether
a role that writes code is still missing a livecode question. Say what you
refreshed and what it changed.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

