Works with the AI Agents you already use
Claude Code
Cursor
Codex CLI
Muse
Dots+22 moreClaude CodeCursorCodex CLIMuseDots
OpenClaw
Manus
Grok
Grok Bot+18 agentsInterview & Case Grader
Grade a completed practice response against a fixed rubric, cite the evidence, and choose the next exercise.
$14.40
Interview & Case Grader
Example session with this skill installed
Realistic evaluator request
Use this package's $interview-case-grader to assess the fictional DockLane written response and mock oral transcript. Return a complete evidence-linked diagnostic assessment and one targeted fresh re-practice. Do not create a new mock world or rewrite the response into a model answer. No real candidate, employer or customer account is involved.
Stage 1 - before response access: Read the packaged candidate case-prompt.md and rubric.md only. Record rubric version, weights totaling 100, written/oral contract, assistance policy, no declared caps, and the rubric hash if file tools are available. Fix this assessment contract before opening the response. The supplied rubric is predeclared by the fixture author; it is not an independently calibrated benchmark. Do not copy it into a blind candidate prompt.
Stage 2 - response review: Then read the packaged candidate-response.md. Apply the fixed rubric. Recompute material arithmetic from supplied facts, cite W/T/I locations, distinguish original written evidence from hinted correction and neutral probes, report coverage, and preserve both weaknesses and strong choices. Do not adjust weights after seeing the response.
Return: main finding and coverage; rubric provenance/read order; severity findings with exact evidence and point impact; earned/available scorecard and score arithmetic; two things to preserve; interviewer-assistance and media limitations; one fresh candidate-facing re-practice with a proposed observable success threshold. Keep any evaluator notes/solution outside the re-practice prompt. Use the user's requested local output destination, or return the assessment in chat if none was supplied. Do not upload, submit, send or predict hiring outcomes. Do not claim fresh independent calibration, an actual timed solve, or voice/delivery evidence from this transcript.
- Read your context and instructions
- Compiled the interview & case
DockLane diagnostic assessment — fictional buyer demonstration
The original written profit argument is not reliable: W3 omits recurring costs and evaluates 80 guided accounts even though the experiment treats 40. The candidate repairs this after a substantive hint and later withdraws premature rollout. Score: 87/100 under the frozen supplied rubric, with 100/100 assessable content coverage. This score describes this fictional attempt, not hiring readiness or an independently calibrated benchmark.
Evidence confidence is high for the cited arithmetic and complete written/transcript record, and moderate for the judgment within scoring bands. Written and oral content are assessed. Vocal delivery, timing, pace, body language, and production implementation are not observed and carry no inferred grade.
Frozen assessment contract — recorded before response access
Rubric
DockLane rubric v1, predeclared by the fixture author from the candidate prompt. The same author supplied the rubric and response; this is not independently calibrated benchmark evidence. This evaluating context has not read the response at contract freeze.
Rubric SHA-256: 4890b10f77f45c9c71bed538e493fe159fc14b0b03d18b5daf6fb43cc22ceaa4.
Read order to freeze: package SKILL.md and evaluator request.md → candidate case-prompt.md → evaluator rubric.md → this contract saved. No source-author QA report or expected answer was read.
| Criterion | Weight | Question / task map |
|---|---|---|
| R1 Decision/customer mechanism | 15 | Written choice and Q1 |
| R2 Evidence and quantitative interpretation | 20 | P1/P5, written evidence, Q1/Q4 |
| R3 Economics: written model | 10 | P2/P3 and written comparison |
| R4 Economics: oral correction/comparison | 10 | Q2/Q3 |
| R5 Experiment and durability decision | 20 | Written test and Q4 |
| R6 Execution, ownership and scope | 10 | P3/P4, written owner and Q5 |
| R7 Oral responsiveness and adaptation | 10 | Q1–Q5 exchanges |
| R8 Communication/format | 5 | Written limit and transcript |
| Total | 100 | Complete written response and five-question oral contract |
Scoring policy is fixed to the supplied anchors and weights. No global caps are declared. Written memo defects remain in R3 even if oral discussion repairs them; R4/R7 can credit visible repair with its assistance level. One defect is not deducted again for interaction unless a distinct unanswered question or unexplained contradiction is observed. Substantive hints and neutral probes are recorded separately. Missing unelicited material is not observed, not automatically zero, and incomplete coverage is not silently normalized to 100. Candidate history is an assertion unless independently evidenced. No employer outcome, hiring-readiness prediction, actual timed solve, voice quality, or vocal delivery will be inferred.
The task is one next-quarter pilot within 80 accounts and six engineer-weeks, a written memo up to 500 words, and five mock oral questions. Either intervention may be defensible under explicit assumptions; recommendation agreement is not a criterion. Decision-driving arithmetic will be independently recomputed from P1–P3. Source omissions, causal effects, power, and longer-term retention remain unknown rather than invented.
The freeze was saved at Stage 1 before response access on October 13, 2026, with the clock showing 13:59:38 UTC. After the saved freeze was read back, the grading references and complete responses/candidate-response.md were read. The rubric hash remained unchanged. Response SHA-256: bd226cca67dd79cdd6e515045f3f6e7f429e4c6fa31cab64ecbc0f14824683bc.
Severity-ordered findings
| Severity / finding | Exact evidence | Decision impact and repair | Point impact |
|---|---|---|---|
| High — written contribution is overstated and footprint differs from the test | W3: “$13,200 incremental contribution”; W4 assigns “40 guided and 40 current-experience accounts.” P2 supplies $40 recurring cost per paid month. | W3's $32,000 is recognized incremental revenue at capacity, not contribution after all costs. At the actual 40-treated footprint and assumed two-month lift, net incremental contribution is −$5,600. Authorize a learning investment, not the original claimed pilot profit. Show recurring contribution, one-time delivery and fixed investment separately at the test size. | R3 3/10, −7. Both linked model defects are assessed here; they are not also deducted as R2 interpretation or R7 interaction failures. |
| High — the written rollout gate can scale an early metric without durable-value evidence | W4: “If the activation threshold is met, I would release guidance to the remaining eligible accounts.” T4 later says “withdraw my earlier automatic rollout rule.” | A first-closure lift is useful learning, not evidence of repeat improvement or additional paid months. The 80-account experiment already assigns everyone; remaining recipients would need an explicit control-crossover/next-cohort decision that preserves the later comparison. T4 improves the rule, but the original authorization remains a written defect. Repair the memo to require a bounded durability/cost review before further exposure. | R5 16/20, −4. The later repair earns substantial combined-test credit; the original gate is not erased. |
| Medium — oral economic repair leaves an actual-size comparison and recovery boundary unfinished | T3: “I have not shown the treatment-size assisted comparison or a precise break-even condition.” | T2 correctly fixes the supplied cost and footprint. T3's alternative-winning inequality is correct for 80 guided versus 40 assisted capacity, but not the common 40-treated pilot. Add the actual-size alternative boundary and a conditional recovery calculation, retaining unknown paid-month lift. | R4 8/10, −2 for incomplete comparison/recovery, not for merely receiving a hint. |
No critical assessment-invalidating defect or declared cap applies. The missing causal paid-month effect, power, and session-time variability are source omissions; refusing to invent them is correct, not a deduction.
Evidence-linked scorecard
| Criterion | Earned / available | Evidence and interpretation | Anchor / confidence |
|---|---|---|---|
| R1 Decision/customer mechanism | 15/15 | W1 makes one constrained guided choice and names correctly authorized closure with reason. W2/T1 treat fields as a hypothesis and preserve responsibility mapping as the counterargument. | Strong; high evidence confidence. |
| R2 Evidence and quantitative interpretation | 20/20 | W2 correctly gives 48/120 = 40% and 30/48 = 62.5%; it disclaims interview prevalence and causal retention. W4 supplies 14-day closure and days 15–28 repeat windows; T4 correctly interprets the hypothetical 20 pp difference without inventing paid-month lift. | Strong; high. W3's cost/footprint defect is scored in R3. |
| R3 Economics: written model | 3/10 | W3 distinguishes effect assumptions, delivery cost and fixed investment, but omits P2 recurring costs and uses capacity instead of W4's 40-treated test. Its positive contribution claim drives the choice. | Weak 0–4 band; high. Later oral repair does not raise the written score. |
| R4 Economics: oral correction/comparison | 8/10 | T2 explicitly acknowledges both errors and correctly recomputes guided capacity, actual pilot, and assisted capacity. T3 correctly gives a capacity alternative inequality, exposes uncertain inputs, and admits missing actual-size comparison/recovery. | Partial 5–8 band; high calculations, moderate point placement. I2 supplies a substantive hint. |
| R5 Experiment and durability decision | 16/20 | W4 has feasible 40/40 assignment, all-assigned denominator including abandonment, valid-closure audit, proposed 15 pp threshold, unauthorized-closure stop, repeat/cost monitoring, and uncertainty. Its original early rollout gate is deficient; T4 explicitly replaces it with bounded follow-up. | Partial 10–16 band; moderate judgment confidence. |
| R6 Execution, ownership and scope | 10/10 | W1 reuses supplied validation/checklist/close/audit and excludes AI/import/bulk scope. W5 names product lead at day 42, engineering logging verification and customer-success cost recording. T5 reduces scope or defers if reliable comparison cannot fit. | Strong; high. These are proposed execution actions, not proven implementation estimates. |
| R7 Oral responsiveness/adaptation | 10/10 | T1–T5 answer all questions. T2 acknowledges error; T3 names remaining repair; T4 changes authorization under the hypothetical; T5 adapts to the instrumentation constraint. | Strong; high for visible content. No second economic deduction or automatic hint penalty. |
| R8 Communication/format | 5/5 | W1–W5 are traceable and concise: 334 whitespace-delimited words after removing W labels, below the 500-word limit. Stable turn IDs make oral reasoning inspectable. | Strong; moderate for communication judgment. No vocal/style claim from prose. |
| Total | 87/100 | 15 + 20 + 3 + 8 + 16 + 10 + 10 + 5 = 87 | Raw = final; no caps. Assessable weight = declared weight = 100. |
The criteria, weights and anchor bands were not changed after response review. No normalized score or hiring band was added. These point assignments are an evidence-linked evaluator judgment within the supplied anchors, not a mechanical measurement of candidate ability.
Arithmetic audit
Source facts are P1–P3; the two-additional-paid-month lift is the candidate's W3 assumption, not an observed intervention effect. All monetary figures below are USD over that hypothetical incremental paid-month horizon. Recognized subscription price is $200 per account-month; ordinary variable cost is $40; recurring contribution is $160 per account-month, or 80% of recognized subscription revenue.
| Conditional model | Treated accounts | Assumed extra paid months per treated account | Recognized extra revenue | Ordinary recurring cost | One-time delivery/service | Fixed pilot investment | Net incremental contribution |
|---|---|---|---|---|---|---|---|
| Guided at capacity | 80 | 2 | $32,000 | $6,400 | $800 | $18,000 | $6,800 |
| Guided actual randomized pilot | 40 | 2 | $16,000 | $3,200 | $400 | $18,000 | −$5,600 |
| Assisted at capacity / hypothetical 40-treated alternative | 40 | 2 | $16,000 | $3,200 | $7,200 | $18,000 | −$12,400 |
Formula: treated × additional paid months × ($200 − $40) − treated × one-time delivery cost − $18,000. Assisted service is 3 × $60 = $180 per treated account; 40 accounts consume the full 120-hour cap. These are separate conditional comparisons of one pilot, not permission to fund both.
For actual 40-treated recovery, guidance requires ($18,000 + $400) / (40 × $160) = 2.875 additional paid months per treated account. Assisted requires ($18,000 + $7,200) / (40 × $160) = 3.9375. Guidance at full 80-account capacity has a different conditional recovery boundary, ($18,000 + $800) / (80 × $160) = 1.46875. None is an observed lift, ROI forecast, or causal estimate.
Let G and A denote the unknown additional paid months per treated account for guided and assisted interventions. At 80-guided/40-assisted capacity, assisted wins when 6,400A − 7,200 > 12,800G − 800, or A > 2G + 1: T3 is correct at that footprint. At a common 40-treated pilot, the comparison is 6,400A − 7,200 > 6,400G − 400, or A > G + 1.0625. Equal fixed investment cancels in each option comparison, but remains in each option's recovery calculation.
P1 first closure is 48/120 = 40%; repeat among first completers is 30/48 = 62.5%. Repeat among all starting accounts would be 30/120 = 25%, a separate denominator that W2 correctly avoids conflating. These are mature cohort associations, not intervention effects. Q4's hypothetical is 28/40 − 20/40 = 70% − 50% = 20 pp; it crosses the proposed 15 pp activation threshold but cannot establish repeat improvement or twelve-month paid-month lift.
Two choices to preserve
Keep the verified-work definition and evidence restraint. W1/W2/T1 focus on a correctly authorized persisted closure with a valid reason, separate checklist activity from value, and preserve the interview counterargument without turning seven convenience interviews into prevalence.
2.
Keep the willingness to narrow or stop. T4 withdraws automatic rollout when repeat use fails to improve. T5 keeps the product lead accountable and defers exposure when reliable comparison cannot fit, rather than treating an uncontrolled rollout as success.
Assistance and limits
| Interviewer turn | Assistance classification | What the candidate demonstrated |
|---|---|---|
| I1 / Q1 | Original question | T1 unpromptedly links valid closures to repeat use and recognizes the alternative obstacle. |
| I2 / Q2 | Substantive hint: explicitly supplies omitted $40 recurring cost and 40-treated footprint | T2 correctly performs and acknowledges the repair. It is assisted correction, not proof that W3 originally used a sound model. |
| I3 / Q3 | Original question / neutral economic probe, with no solution supplied | T3 gives a correct capacity boundary and admits the missing actual-size/recovery work. |
| I4 / Q4 | Original declared changed-premise question, not a supplied interpretation | T4 calculates the rate difference and withdraws the earlier automatic authorization. |
| I5 / Q5 | Original declared resource-constraint probe, not an answer hint | T5 narrows scope conditionally and refuses unverifiable rollout. |
No model-answer coaching or extra unrecorded probes are visible. All written sections and five oral turns are supplied and declared complete; no scored criterion is marked unobserved. Audio/video, actual elapsed solving time, pause patterns, delivery, statistical power, causal paid-month effects, and implementation estimates remain unavailable. The transcript supports content/adaptation judgments only. The record cannot establish that a real solver was blind or independently timed; it is an author-supplied fictional diagnostic demonstration.
One fresh targeted re-practice
The next attempt targets the original economic authorization weakness without giving the corrected calculation. Copy only the candidate-facing block below for practice; the audit above and evaluator comments remain separate.
Fresh DockLane constraint — candidate-facing practice. All numbers are fictional. You have 60 eligible accounts and may choose one pilot with 30 treated and 30 current-practice accounts. The monthly recognized price is $180 and ordinary recurring variable cost is $50 per paid account-month. Fixed pilot investment is $15,000. Guidance costs $12 once per treated account. Assisted mapping costs two hours at $60/hour per treated account and has 60 service hours available. Assume guidance creates 1.5 additional paid months per treated account for a conditional comparison; assisted lift is unknown. No causal lift or long-term paid-use result is supplied.
Write at most 200 words comparing the actual treatment-size economics, a conditional recovery boundary, and an assisted-winning condition. Make a bounded authorization decision and state which effect/cost inputs remain assumptions. Then orally defend the actual-treatment contribution and the alternative-winning condition, and answer this neutral probe without a substantive hint: “If first verified closure improves but repeat use does not, what do you authorize next and why?” Do not replace the 30 treated accounts with total eligible capacity or claim that activation proves paid-month lift.
Proposed observable success threshold: all decision-driving arithmetic is correct; recognized revenue, recurring contribution, delivery and fixed investment remain distinct; the actual treatment denominator, one conditional recovery boundary, one alternative-winning condition and a durability-aware authorization rule are explicit. Target at least 19/20 across R3/R4's unchanged anchors without a substantive hint, with no unsupported observed-effect claim. This narrow exercise assesses 20/100 of the original contract; the durability response is an additional qualitative guardrail, not a separate R5 rescore. Other criteria are not fully elicited and must not be silently scored or normalized. No re-practice has yet been executed, and improvement on this fresh constraint would support transfer of the model discipline rather than an interview-outcome prediction.
Connects securely to your tools. The creator never sees your data.
About this skill
What it helps you do
Assess a completed practice case, interview transcript or oral defense against a rubric fixed before response review. Preserve exact response evidence, distinguish unprompted reasoning from repairs after hints, and turn substantiated gaps into a targeted next exercise.
What is included
- A focused SKILL.md assessment workflow.
- Rubric and evidence rules, oral-interview assessment guidance, and a feedback/re-practice report contract.
- An original fictional case, predeclared evaluator rubric and practice response, kept in separate role-specific files.
How to use it
Supply the task, role level, rubric and frozen response. Follow the declared read order. The agent maps each assessed criterion to response locations, checks decision-driving arithmetic, distinguishes an observable omission from an untested competency, and reports earned points alongside assessment coverage. Missing modalities stay not observed.
How it complements the practice tools
Executive Case Practice Studio creates cases and evaluates its generated practice. This grader focuses on supplied written and interview responses, including interviewer assistance, partial evidence and observable adaptation. It does not create a full mock case or replace the timed coach.
Boundaries
Diagnostic practice feedback does not predict hiring outcomes. A transcript cannot establish voice, pacing, body language or personality. Candidate career assertions remain unverified unless supporting evidence is supplied. If this context already saw the response, the assessment must disclose exposure rather than claim a blind rubric. Separate files do not prove role isolation. Live assignments must permit the chosen assistance.
How to use
Skills on Agensi run inside your AI agent. Copy the prompt, paste it into your agent, and it connects to Agensi and installs this skill for you. If a payment is needed, your agent shows it to you first.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scanned
Verified clean 6 days ago
- Passed all security checks, Safe to install