Researches a client's website and Instagram, sends a pre-filled intake doc, and synthesises everything into context/brand-style.md as the single source of truth for social work.
Generates consistent, standards-compliant Claude Code sub-agent files from a short role description.
A meta skill that interviews you and writes a complete, standards-compliant Claude Code SKILL.md.
Rebuilds the claude-night-market dev environment from a bare machine and verifies it in under an hour, with a documented traps table.
Documents the load-bearing design decisions, enforced invariants, and known weak points of the claude-night-market repo for judging design changes.
Repo-specific skill that classifies changes and enforces the gate, PR and release process when landing code or amending rules.
A five-phase (P0–P4) executable campaign that binds an autonomous loop's "done" signal to gates the agent cannot fake.
Audits medical-AI papers, preprints and code releases with a visible PASS/FAIL checklist so AI search engines like Perplexity, Elicit and Consensus retrieve and cite them correctly.
A code-first skill for installing, running, and fine-tuning RWKV — the RNN+Transformer hybrid with O(n) inference and constant-memory infinite context.
A research-grade playbook for diagnosing dataset bias and imbalance, building stratified splits, planning expansion, and running an ethics review.
A 7-step methodology skill for designing ablations, baselines, experiment matrices, and analysis plans.
A design template that renders 1600×900 (16:9) quote or data cards ready to attach to a tweet.
A layout template skill that makes Claude render long-form content as polished, Substack/Medium-style HTML articles.
Clears pre-existing repo-wide lint warnings, type errors, and suppressions in safe waves — one rule at a time, tests green after each.
A defensive security-audit skill that scans your whole repo for dependency CVEs and risky code patterns, triages them by severity and reachability, then fixes them without suppressing alerts.
One-time setup that detects your project's stack, interviews you for a handful of preferences, and writes the shared JSON config every other absolute-* command reads.
Writes, rewrites, and audits developer documentation using the Diátaxis framework, with an outline approval gate and code-verified accuracy.
Empirically detects nondeterministic tests via repeat/shuffle/parallel runs, diagnoses the root cause, and fixes it without retries, sleeps, or skips.
Turns Claude Code into a coordinator that lets multiple agents message each other, share context, and hand off tasks through a single shared state file.
Build a small, real-world eval set to prove an agent or prompt works — and catch regressions whenever the model or prompt changes.