execution: 154/160 graded | format 3.55 method 4.13 usefulness 4.14 | grounded 134/154
  LOW weekly-review                    f1 m1 u1 grounded=False | Does not produce the required output; Does not follow workflow; Hallucinates capabilities; Acts as a conversational chatbot
  LOW inbox-triage                     f1 m1 u1 grounded=True | Does not follow output format; Does not apply skill workflow; Fails to acknowledge missing data; Generic response
  LOW meeting-prep                     f1 m1 u1 grounded=True | Returned a technical error message instead of an answer.; Did not follow the skill's workflow.; Did not ask for the required meeting information.; Did not state that data sources were unavailable.
  LOW follow-up-tracker                f1 m1 u1 grounded=False | Falsely claims it will check connected apps; Does not acknowledge the lack of data sources; The answer is incomplete and provides no actual information
  LOW back-from-leave-catchup          f1 m1 u1 grounded=True | Returned a technical error instead of an answer.; Did not follow the skill instructions.; Did not explain the failure or ask for needed information.
  LOW team-newsletter                  f1 m1 u1 grounded=True | Did not produce requested output; Did not follow skill instructions; Fails to mention lack of data sources
  LOW account-research                 f1 m1 u1 grounded=True | Returned a technical error; Did not produce any output
  LOW b2b-proposal-builder             f1 m1 u1 grounded=True | Did not follow instructions; Did not use the specified output format; Failed to apply skill workflow; Pushed work back to the user instead of creating a draft
  LOW topic-digest                     f1 m1 u1 grounded=True | Is an error message; Did not follow output format; Did not apply skill; Did not handle unavailable sources
  LOW release-readiness-check          f1 m1 u1 grounded=True | Did not follow workflow; Did not use output format; Did not mention unavailable sources; Blocked on optional input
  LOW daily-briefing                   f1 m1 u1 grounded=True | Filler response; Did not execute skill
  LOW new-manager-90-day-plan          f1 m1 u2 grounded=True | Did not generate the plan as requested.; Failed to follow the specified output format.; Asked for optional inputs instead of generating a plan with defaults and assumptions.; Did not follow the skill'
  LOW onboarding-plan-builder          f1 m1 u2 grounded=True | Did not generate the plan as requested; Failed to follow the specified output format; Asked for optional information (manager, contract type, JD) instead of using placeholders or defaults; Did not app
  LOW competitive-battlecard           f1 m2 u1 grounded=True | Did not follow the output format; Failed to generate the requested deliverable; Could have provided an empty template instead of refusing
  LOW expert-finder                    f1 m2 u1 grounded=False | Does not follow output format; Incomplete response; Pretends to search unavailable sources
  LOW backlog-grooming                 f1 m2 u1 grounded=True | Does not produce the required output format; Fails to follow instructions for unavailable sources; Asks for permission instead of stating the source is unavailable; Does not produce a final answer
  LOW prioritization-coach             f1 m2 u2 grounded=True | Did not follow output format; Did not complete workflow; Puts work back on user
  LOW okr-builder                      f1 m2 u2 grounded=True | Did not follow output format; Did not state inability to access source; Did not continue with available information
  LOW csr-data-request                 f1 m2 u2 grounded=True | Does not follow output format; Does not apply skill workflow; Could have produced a partial result with placeholders
  LOW store-visit-report               f1 m2 u2 grounded=True | Does not generate the report; Too conversational; Fails to use provided notes
  LOW product-page-copy                f1 m2 u2 grounded=True | Did not use the specified output format; Did not follow the workflow to generate a skeleton with placeholders; Returned a generic conversational refusal instead of applying the skill
  LOW sprint-planning-assistant        f1 m2 u2 grounded=True | Asks for source access it doesn't have; Fails to ask for the data directly; Leads the user to a conversational dead-end
  LOW data-question-answerer           f1 m2 u2 grounded=True | Does not follow output format; Does not ask for missing required inputs
  LOW email-reply-drafter              f1 m2 u3 grounded=True | Does not use the skill's output format; Fails to state that a source is unavailable
  LOW workshop-designer                f1 m2 u3 grounded=True | Fails to generate the workshop plan as requested.; Asks a clarifying question instead of making a reasonable assumption for the outputs and generating a full draft plan.; Does not follow the specified
  LOW offer-letter-drafter             f1 m2 u3 grounded=False | Does not use the specified output format; Falsely implies it can search Drive; Does not state that data sources are unavailable; Asks questions instead of generating a draft with placeholders
  LOW store-incident-report            f1 m2 u3 grounded=True | Did not generate the report in the specified markdown format.; Failed to follow the skill's instructions for handling missing information, which is to generate a partial report and list questions with
  LOW okr-progress-review              f1 m4 u2 grounded=False | Pretends to have access to data sources; Sets false user expectations
  LOW ai-use-case-finder               f1 m3 u4 grounded=True | Does not follow output format; Incomplete workflow; Asks for information instead of providing a full answer based on assumptions
  LOW meeting-notes-to-actions         f5 m2 u2 grounded=False | Does not state that the source (Drive) is unavailable, as required by the skill.; Implies it can search the user's Drive for a file if given the name, which is false.; Asks for information (a filename
  LOW decision-memo                    f5 m2 u2 grounded=True | Recommendation contradicts scoring; Mixed languages in output
  LOW self-assessment-coach            f4 m3 u2 grounded=False | Pretends to have access to data sources
  LOW pr-review-assistant              f1 m4 u4 grounded=True | Does not use the skill's output format
  LOW project-plan-builder             f5 m2 u2 grounded=True | Critical path calculation is wrong; Contradictory end date; Incorrect feasibility analysis
  LOW process-improvement-kaizen       f1 m4 u5 grounded=True | Does not follow output format; Asks clarifying questions instead of generating a templated analysis
  LOW translation-localizer            f1 m5 u5 grounded=True | Does not use the skill's output format, but this is appropriate as a required input (the source text) was missing.
  LOW sales-call-recap                 f5 m3 u3 grounded=False | Invented Airtable schema; Did not state source unavailability
  LOW customer-reply-drafter           f5 m3 u3 grounded=False | Invents policy details; Contradictory placeholder list
  LOW gdpr-processing-check            f1 m5 u5 grounded=True | did not generate the required output format; did not attempt a partial analysis with placeholders
  LOW regulatory-watch-digest          f1 m5 u5 grounded=True | 
  LOW email-newsletter-writer          f1 m5 u5 grounded=True | Does not generate the final output, but correctly identifies missing required inputs and asks for them before proceeding.
  LOW marketing-performance-report     f1 m5 u5 grounded=True | Did not produce the final report format, but correctly asked for the required inputs first, which is the appropriate action.
  LOW release-notes-writer             f1 m5 u5 grounded=True | Did not use the skill's output format, instead asking for the required inputs which were missing from the prompt.
  LOW dashboard-spec-writer            f5 m3 u3 grounded=False | Invented source table names; Invented data latency and refresh rates
  LOW cross-source-search              f5 m5 u1 grounded=False | invents facts; invents sources; falsely claims to have searched sources; hallucinates content
  LOW engineering-standup-summary      f1 m5 u5 grounded=True | Does not produce the requested output, but correctly asks for the necessary inputs to do so.
  LOW sop-writer                       f5 m3 u3 grounded=True | RACI chart is missing the 'Accountable' role for several activities, violating a quality check.; A decision point in the procedure has incorrect logic (step 6 directs to the wrong subsequent step).; O
  LOW budget-variance-analysis         f1 m5 u5 grounded=True | 
  LOW meeting-agenda-builder           f5 m3 u4 grounded=True | Did not shorten meeting from 60 to 50 minutes as per skill instructions.; Agenda has no buffer time; skill requires a 10% buffer.; Discuss items are 8 minutes long, shorter than the 10-minute minimum 
  LOW store-action-plan                f5 m4 u3 grounded=False | Invented specific numbers (e.g., '30% des références', 'temps d'attente > 5 minutes pour 20% des clients', '40% des cabines') for the problem statements, presenting them as facts from an audit that wa
  LOW customer-policy-qa               f3 m4 u5 grounded=True | Does not follow the final output format, but this is an intermediate clarification question.; Does not restate the question, which is the first step in the workflow.
  LOW customer-persona-builder         f5 m4 u3 grounded=False | invented evidence; invented quotes
  LOW experiment-design                f5 m3 u4 grounded=True | Primary metric definition (view-based) mismatches the randomization unit (user-based).; The primary metric definition in the main table contradicts the (correct) user-level metric calculated in the ex
  LOW skills-navigator                 f3 m4 u5 grounded=True | Output format not strictly followed; Recommended more than the 1-3 skills requested
  LOW document-reviewer                f3 m5 u5 grounded=True | Slightly redundant question about user's concerns, as the user already mentioned 'logique, trous, clarté'.
  LOW feedback-sbi-writer              f5 m4 u4 grounded=False | Invented impact instead of asking the user as instructed
  LOW performance-review-writer        f3 m5 u5 grounded=True | 
  LOW career-development-plan          f3 m5 u5 grounded=True | Does not produce the final output format, but correctly follows the workflow by first asking for the required 'aspiration' inputs.
  LOW team-skills-matrix               f3 m5 u5 grounded=True | 
  LOW exit-interview-synthesizer       f3 m5 u5 grounded=True | 
  LOW learning-path-designer           f3 m5 u5 grounded=True | Did not generate the learning path as requested, but instead asked for the required inputs defined in the skill.
  LOW prospect-outreach-sequence       f5 m4 u4 grounded=False | Invented a specific metric ('20%') without a source
  LOW account-plan                     f3 m5 u5 grounded=True | 
  LOW win-loss-analysis                f3 m5 u5 grounded=True | 
  LOW sales-forecast-commentary        f3 m5 u5 grounded=True | 
  LOW store-performance-analysis       f3 m5 u5 grounded=True | 
  LOW sport-universe-review            f3 m5 u5 grounded=True | 
  LOW seasonal-operation-planner       f3 m5 u5 grounded=True | 
  LOW ticket-triage                    f3 m5 u5 grounded=True | 
  LOW kb-article-writer                f3 m5 u5 grounded=True | 
  LOW voice-of-customer-report         f3 m5 u5 grounded=True | Slightly redundant question about which sources to include, as the user already listed them.
  LOW product-review-insights          f3 m5 u5 grounded=True | 
  LOW response-template-library        f3 m5 u5 grounded=True | Does not explicitly state that data sources are unavailable to find the top contact reasons, but correctly asks the user for them instead.
  LOW campaign-brief                   f3 m5 u5 grounded=True | 
  LOW competitor-watch                 f3 m5 u5 grounded=True | Does not explicitly state that internal sources (Drive, Gmail...) are unavailable, as instructed by the skill.
  LOW field-test-protocol              f5 m4 u4 grounded=False | Invents specific pass criteria; Invents a specific name for the reference product
  LOW competitive-product-teardown     f3 m5 u5 grounded=True | 
  LOW eco-design-review                f3 m5 u5 grounded=True | 
  LOW tableau-dashboard-explainer      f3 m5 u5 grounded=True | 
  LOW insight-storyteller              f3 m5 u5 grounded=True | 
  LOW glossary-builder                 f3 m5 u5 grounded=True | 
  LOW bug-report-writer                f5 m4 u4 grounded=False | Pretended to have checked for duplicates in Jira and found none, instead of stating the source was unavailable.
  LOW incident-postmortem              f3 m5 u5 grounded=True | 
  LOW business-case-builder            f3 m5 u5 grounded=True | 
  LOW supplier-evaluation              f3 m5 u5 grounded=True | Did not explicitly state that internal data sources were unavailable before asking the user for information.
  LOW purchase-request-justification   f3 m5 u5 grounded=True | 
  LOW price-positioning-analysis       f3 m5 u5 grounded=True | 
  LOW monthly-finance-commentary       f3 m5 u5 grounded=True | 
  LOW kpi-definition-card              f5 m5 u4 grounded=False | Invented a specific source table name (`company_db.repairs.repair_events`) despite having no access to data sources.; Invented specific column names (`repair_id`, `repair_start_timestamp`, etc.) and p
  LOW faq-builder                      f5 m5 u5 grounded=False | Invents specific details (policy name, URL, section numbers, dates, prices) instead of using placeholders.
  LOW architecture-decision-record     f5 m5 u5 grounded=False | Invented a specific date instead of using a placeholder
