Run3 evidence validation checkpoint

Skill cxcscmu/SkillLearnBench/skills/b3-teacher-feedback-claude-haiku-4-5/enterprise-information-search/run3_evidence-validation-checkpoint

[COLM'26] SkillLearnBench is the first benchmark for evaluating continual learning methods that automatically generate agent skills.

Install

npx -y skills add cxcscmu/SkillLearnBench --skill run3_evidence-validation-checkpoint

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

What its author says it does

Copied from the file, not written here

Before writing final answers, validate that all required evidence has been extracted, multi-hop traversal was executed, and answers are complete against expected values.

SKILL.md

1.8 KB, 315 tokens by cl100k_base, as published. Nobody here has run it

Evidence Validation Checkpoint Skill

Step 1: Verify Tier Extraction Completeness

Confirm Tier 1 (explicit reviewers) extracted from all artifacts
Confirm Tier 2 (Slack/transcript contributors) extracted with substantive feedback filters applied
Confirm Tier 3 (other contributors) identified
Document what was found in each tier

Step 2: Validate Multi-Hop Traversal Execution

Confirm traversal was actually performed (not skipped)
Verify all artifact references were followed
Confirm reviewers from referenced artifacts were included
Print the traversal paths taken

Step 3: Check for Missing Items

For each question, compare extracted answers against known required items
Identify any missing names/IDs
Trace back to find where each missing item should have been extracted from
Re-extract if needed

Step 4: Validate Meeting Participant Rule

Review all extracted names from meeting transcripts
Confirm each person actually spoke and provided substantive feedback
Remove any names that were only listed in participants arrays
Document any removed names and why

Step 5: Deduplicate Final Answers

Remove duplicates within each question's answer list
Maintain alphabetical or logical ordering if applicable
Verify list format is correct (always a list, never string)

Step 6: Confirm Before Writing

Print final answers for all questions
Verify answer counts are reasonable (not zero when expected to have results)
Confirm token counts are numeric values
Only proceed to JSON output if all validations pass