Claude Code craft for working engineers: authoring Agent Skills (the library's own standard), prompt engineering with the injection-defense doctrine, git/code review with run-verified agent traps, the Board of Advisors multi-agent code review (board-review + five specialist subagents and a chair), defect-epidemiology (treat a confirmed bug as evidence about a population), rule-stress-testing, software-archaeology, Comrade Engineer's pencil pass, The Foreman's draw inspection, the Creative Workaround Genius (another route when the obvious one fails, never around a control), the project-manager subagent with its skill for multi-session and multi-agent work (requirement ledger, token budget, exhaustive QC sweep), Puppet Master (the whole job on the fewest tokens with no loss of quality: fewer turns, small new context, a transcript audit), and Chicken Little (Aether) — the multi-domain persona with forward-deployed advisor and compiler autopsy modes.
13 skills · 7 agents
Decision discipline under uncertainty: pre-mortem (with the counsel rail), tabletop wargaming and no-win drills under full exercise control, weak-signal navigation with cue-graded confidence, Bayesian updating via natural frequencies, reference-class forecasting, competing hypotheses analysis, minority-report dissent capture, the Rashomon-effect protocol for conflicting recollections, and The Challenger.
10 skills
Full-stack application development with a lean-code philosophy: architecture, FastAPI backends, databases/ORM, modern dynamic frontends, realtime features, ML in production, testing strategy, deployment, and evidence-based UI/UX inspection with severity-rated findings and human-factors instruments (Fitts, NASA-TLX).
10 skills
The improvement methods with teeth: DMAIC with a fully worked control-chart project (provisional-limits discipline included), FMEA with the evidence-gated Action Priority table, measurement systems analysis extended to LLM-as-judge gauge studies, design of experiments, evolutionary operation (EVOP), root cause analysis, and the project command center doctrine.
7 skills
Statistical inference done honestly, causal inference (DAGs, DiD/IV/RDD), A/B test design (SRM, peeking, MDE, variance reduction), survey and sampling design (total survey error, nonresponse), spreadsheet modeling discipline, and assertion-evidence presentations with a build-and-lint toolchain.
6 skills
Applied machine learning with honest evaluation: project framing (the leanest model is no model), supervised modeling, feature engineering, model evaluation (threshold as a fitted parameter), time-series forecasting with horizon-aware backtest sizing and measured interval coverage, and anomaly detection with the degenerate-case floor.
6 skills
Safety engineering for records, alarms, and dormant protections: split-tally evidence design with the litigation-hold rail, detection-system tuning by empirical disposition audit (floor-ceiling autoimmunity band), reliability math including probability-of-failure-on-demand for dormant protective functions, rebuild rehearsal with an irreversibility screen, break-glass playbooks, and hash-rank sortition review.
6 skills
Hands-on data plumbing that survives real files: DuckDB local analytics with a null-key-guarded reconciliation pattern, PDF table extraction hardened against full-width headings and dateless rows, REST API pulls with bounded pagination and per-request credentials, CSV/flat-file wrangling, and data-file hygiene.
5 skills
True deep research skills that go beyond surface search: extensive multi-database literature investigation, cross-domain dot-connection, source-provenance control, evidence appraisal, and citation verification. Covers published medical literature (medical-research-detective), auditing video claims against the primary sources they cite with A+ to F grading (video-source-audit), and interpreting personal genome raw data against the literature with absolute-risk framing and hard clinical-confirmation rails (dna-raw-data-audit). Also turns the same genome into an evidenced ancestry story — open-tool haplogroup calls, proof-standard genealogy and lineage maps (dna-ancestry-research).
4 skills
Writing registers and explanation craft: adams-smart-brevity (Adams + Axios Smart Brevity for professional, technical, legal, and clinical writing, with the risk-allocation carve-outs and the where-the-bar-disagrees register), adams-plain-grade (5th-8th grade accessible register with the discretion/obligation check), explanation-design (audience models, break-point-marked analogies, the Feynman loop, teach-back), and technical-documentation (Diataxis routing, ADRs, and the owner/cadence/last-verified discipline that keeps a page true).
4 skills
Legal drafting, research, and employment-law issue-spotting with Smart Brevity delivery: mscd-contract-drafting (Ken Adams's A Manual of Style for Contract Drafting applied as a deterministic linter plus a redline protocol — categories of contract language, shall discipline, defined terms, efforts, provisos, ambiguity — with flag-don't-cut rails for risk allocation and prescribed wording), legal-citation-research (research plan, authority hierarchy, no-fabricated-citation rail with offline citation parsing and CourtListener verification that fails to UNVERIFIED, never to a false verified), and employment-law-issues (HR issue-spotting across the employee lifecycle with the live-source verification discipline, because the thresholds move). Every memo, redline cover note, and issue list leads with the point and why it matters.
3 skills
Composed meta-cognition suite: hierarchical memory management, reflective learning, and knowledge crystallization for cumulative improvement across sessions.
3 skills
Two collaboration instruments that earn their slots: disarming-elicitation (the FBI counter-elicitation stance inverted for willing experts, with the scoped-promise and investigation rails) and meeting-design (decisions extracted as owner-and-date read-backs, not vibes).
2 skills
Unit conversion and dimensional analysis — the factor-label method, unit checks as a necessary-not-sufficient guard on every formula, and the unit-driven catch for formula rearrangement errors. The broader basic-math set is archived; this is the one that catches real errors models and humans both make.
1 skill