lyndonkl/claude
6 skills1.2k installs828 starsGitHub
Install
npx skills add https://github.com/lyndonkl/claudeSkills in this repo
1Evaluation Rubricsevaluation-rubrics is an agent skill from lyndonkl/claude aimed at developers building or maintaining AI agents who need consistent quality measurement. The skill guides creation of evaluation rubrics that score agent responses, tool use, and task completion against defined criteria instead of relying on subjective spot checks. Developers reach for evaluation-rubrics when setting up agent QA, regression comparisons, or prompt iteration loops where repeatable pass-fail and graded scoring matter. Output is a reusable rubric framework teams can apply across test cases, eval runs, and release gates.629installs2Brainstorm Diverge Convergebrainstorm-diverge-converge is an agent skill implementing a five-step diverge-cluster-converge workflow for product ideas, problem-solving, research questions, and strategic planning. Step 1 gathers requirements—topic, goal, constraints, evaluation criteria, and a 20–50 idea target. Step 2 diverges by generating 20–50 ideas without judgment using creative prompts. Step 3 clusters ideas into 4–8 distinct theme groups. Step 4 converges by scoring against explicit criteria (impact, feasibility, cost, speed, risk, alignment) and selecting the top 3–5 options with documented tradeoffs. Step 5 writes `brainstorm-diverge-converge.md` and validates against `resources/evaluators/rubric_brainstorm_diverge_converge.json` requiring score ≥ 3.5. Bundled `resources/template.md` supports complex topics. Reach for this skill when exploring open-ended decisions—not when executing a fixed spec or writing production code.126installs3Decision Matrixdecision-matrix is a Claude agent skill from lyndonkl/claude that walks developers through weighted multi-criteria decision analysis when choosing between concrete named options such as vendors, tools, hiring candidates, or feature priorities. The workflow has five steps: frame the decision and list alternatives, identify criteria and assign percentage weights totaling 100%, score each option on a 1–10 scale, calculate weighted totals with sensitivity checks, and deliver a decision-matrix.md file validated against rubric_decision_matrix.json at a minimum average score of 3.5. The skill documents four weighting methods—direct allocation, pairwise comparison, must-have filtering, and stakeholder averaging—and flags fragile decisions when winners change within 5–10% margins or small weight shifts. Bundled references include template.md, methodology.md, and rubric_decision_matrix.json. Reach for decision-matrix when balancing cost versus quality versus speed, evaluating vendors, or facilitating group decisions that need visible trade-offs instead of gut-feel picks.126installs4Design Of Experimentsdesign-of-experiments is a Claude agent skill from lyndonkl/claude that helps developers plan statistically rigorous experiments when optimizing systems with multiple controllable factors under run-budget constraints. The six-step workflow defines objectives and constraints, lists factors levels and responses, selects a design type, plans randomization and replication, writes design-of-experiments.md with the run matrix, and validates against rubric_design_of_experiments.json at a minimum average score of 3.5. It covers five common patterns: screening 10–30 factors via Plackett-Burman or fractional factorial, optimizing 2–5 factors, mapping response surfaces with central composite or Box–Behnken designs, Taguchi robust designs against noise factors, and sequential screening-to-confirmation flows. Guardrails require randomized run order, replicated center points, Resolution IV or higher when interactions matter, correlation below 0.3, and power analysis targeting beta at most 0.20. Reach for design-of-experiments when tuning software performance parameters, running A/B/n experiments, or replacing one-factor-at-a-time tuning that misses interactions.124installs5Career Document Architectcareer-document-architect is an agent skill from lyndonkl/claude that structures academic job and promotion writing through a 7-step workflow: identify document type and audience, gather materials, develop a core narrative, draft with type-specific frameworks, add quantified evidence, align with institutional expectations, and polish against a JSON rubric (minimum average score ≥ 3.5). It covers research statements (typically 2–5 pages), teaching statements (1–2 pages), diversity statements (1–2 pages), academic CV sections, and NIH biosketches (5 pages maximum with up to 5 Contributions to Science). Frameworks emphasize vision plus track record, the “So What?” test, and showing versus telling with citations and mentorship counts. Related skills handle grants (`grant-proposal-assistant`) and letters (`academic-letter-architect`). Reach for it when drafting or revising faculty applications, promotion packages, fellowship statements, or biosketches—not for scientific manuscript peer review.115installs6Expected Valueexpected-value is a Claude Code skill for ai & agent building. It helps solo builders move faster with AI-assisted development.103installs