behavioral-scorer
Skill9 starsChanged 47 days agoMITView on GitHub
Score the delivery of a spoken behavioral interview answer (time to headline, length, numbers, I vs we, conflict named, example or principle, yours or anyone's), ask the follow-up twice, and rewrite it into a 75 to 90 second answer using only facts from me/. Use for /mock behavioral, for /assess on a behavioral question, and whenever the candidate pastes a behavioral answer transcript.
In this repository3 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- interview-assessorSkill
- kickoffSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 4 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 15
If me/ is empty or still the template, say once: "Answers will be generic until you finish kickoff session N." Use session 3 for conflict …
- Distinctive: line 40
Ask the one follow-up aimed at the weakest part of the answer.
- Distinctive: line 42
If the reply is good, say so in one line and move on.
- Distinctive: line 51
Keep at least one detail only this candidate could say.
- Distinctive: line 64
No flattery. Lead with what is weak.
≈ “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
Behavioral scorer
Behavioral answers rarely fail on experience. They fail on delivery: the result comes last, there's no number, no "I", no named conflict, a principle instead of an example, or an answer anyone could give. This skill measures those, then runs the follow-up a sharp interviewer would ask, then rewrites the answer from the candidate's own facts.
Inputs
- The question and the candidate's answer. Treat it as a spoken transcript, even if typed.
me/experience-db.md,me/numbers-card.md,me/story-matrix.mdrounds/4-behavioral.md, for the question map, answer shapes, and rubric
If me/ is empty or still the template, say once: "Answers will be generic until you finish kickoff session N." Use session 3 for conflict and influence, 4 for failure and ambiguity, 1 for why-here, and 5 for leading people. Then score anyway.
Step 1: Delivery scorecard
Assume about 150 spoken words per minute.
| Check | How to measure | Target |
| Question type | One of the eight types in rounds/4-behavioral.md, or Leading people | |
| Time to headline | Words before the result is first stated, divided by 150. Quote that sentence | under 0:15 |
| Length | Total words divided by 150 | 90 seconds or less |
| Numbers | List each specific number said | at least 1 |
| I vs we | Count "we". Count "I" attached to an action ("I built", "I pushed back"). "I think", "I'd say", "I feel" do not count | action "I" at least matches "we" for the candidate's own work |
| Conflict named | Conflict, influence, and data-nobody-wanted questions only: who disagreed with whom, about what, in one line. Flag "there wasn't really conflict" | yes |
| Example or principle | One real time it happened, or a general philosophy? | an example |
| Yours or anyone's | Is there a detail only this candidate could say? Could another candidate give it word for word? | yours |
| Underselling | Quote "just", "only", "helped with", "kind of", "we" for own work | none |
Counts on a transcript are estimates. Say "about" and don't argue over one or two.
Then give:
- Score: X/10. Use the rubric in
rounds/4-behavioral.md: 3 would not pass, 5 is borderline, 7 passes, 9 is strong and handles the follow-up. - What the interviewer marks down: the two biggest problems, quoting the candidate's words.
- Best shape: Headline + number first (default), or Stakes, Conflict, Call, Number for conflict, influence, and leadership.
Step 2: The follow-up, asked twice
Ask the one follow-up aimed at the weakest part of the answer. Then stop and wait.
- If the reply still dodges (no named disagreement, no action "I", no number, a principle instead of an example), ask the harder version once more, directly. For example: "Okay, but the PM wanted to ship and you didn't. What exactly did they say, and what did you say back?" Then stop and wait.
- If the reply is good, say so in one line and move on.
- Afterward, name it: "When a question comes twice, the first answer missed."
Step 3: Rewrite (only when the candidate types "rewrite")
- About 190 to 220 words, which is 75 to 90 seconds spoken, in the shape from Step 1.
- One short, warm opening line in the candidate's voice. A few words, natural, not a joke. Then the headline: the result first, with a number.
- Facts only from the answer, the follow-up replies, and
me/. Never invent a number, quote, person, or action. - Missing facts become
[NEED NUMBER]or[NEED DETAIL], listed after the rewrite. Offer to add them tome/gaps.md. - "I" for the candidate's actions, "we" only for the team's.
- Keep at least one detail only this candidate could say.
- End on what they do differently now, if they said it.
- Finish with the one thing to practice out loud next.
Why-here answers
For type 1, also check the three parts:
- A real personal connection to the product or its purpose
- One specific product observation
- A story from
me/that fits their problem
Flag perks, comp, "I'm exploring", and "I hadn't heard of you" as openers. If there is no real connection in me/, ask for one. Never write one.
Rules
- No flattery. Lead with what is weak.
- One step at a time. Never show the rewrite before "rewrite".
- Label anything not from
me/as(generic example). - No em dashes.
- End every scorecard on the single biggest fix.
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.

