Generate a structured markdown audit document for metrics or pipeline steps.
npx skills add https://github.com/vitalwarley/hyprdots --skill audit-vitalwarley
Please help me install this Agent Skill. Skill: audit Source: https://github.com/vitalwarley/hyprdots/tree/main/claude-global/skills/audit Command: npx skills add https://github.com/vitalwarley/hyprdots --skill audit-vitalwarley
This Skill helps researchers and engineers verify that metric implementations and pipeline steps are correct, reproducible, and appropriately documented so that reported values in analyses and papers can be trusted.
Audit the metric gaussian entropy by locating its implementation, recomputing results against the canonical embeddings, and writing a reproducible audit document to report/sprints/week-NN/gaussian-entropy-audit.md.
High-intent search queries and answers about installing and using this skill.
Pipeline auditing involves reading code and reports, verifying embedding provenance and checkpoint validity, recomputing metrics with reproducible scripts, and producing a linked markdown document with evidence, findings, limitations, and a safe claim statement.
A reproducible audit report for a pipeline step includes pre-conditions, process verification checks, post-conditions, findings, limitations, a safe claim statement, and references to commits and evidence for full traceability.
After completing a pipeline audit, you create issues for investigation, update global audit trackers, and reference commits and evidence for traceability to maintain governance over the codebase validation process.
Locating a metric implementation requires reading the research codebase to find the function or script computing the named metric, then enumerating its inputs, outputs, and assumptions to verify correctness before recomputing values.
Code review and revalidation auditing differ from standard code review by requiring reproducible scripts to recompute metric values against specified embedding files, capturing diagnostics, and comparing them to reported numbers in a formal markdown document.
Limitations of revalidation auditing include dependency on reading existing code and reports, requiring specified embedding files and validated checkpoints, and the need for reproducible scripts to ensure accurate recomputation and comparison of metric values.
drpedapatiRecord experiment provenance and reproducible run metadata for research workflows.
KentoShimizuTranslate validated elicitation evidence into a canonical, testable requirement baseline with traceability and ownership mapping.
CodingHeaderRun verification commands and compare output against expectations before claiming completion.
frogguguguguReview code changes for quality, conventions, performance, and security.
