Skip to main content
KurtLehnardt
Perfil de criador do GitHub

KurtLehnardt

Visão por repositório de 19 skills coletadas em 1 repositórios do GitHub.

skills coletadas
19
repositórios
1
atualizado
17 de ago. de 2026
explorador de repositórios

Repositórios e skills representativas

survive
Desenvolvedores de software

Wrap a worker loop so smaller or local models actually finish — response validation, rescue parsing, corrective nudges, step enforcement, and context compaction. The model proposes the next action; the harness makes the loop survivable. Trigger: make the loop…

17 de ago. de 2026
observe
Desenvolvedores de software

Instrument an agent or LLM feature with structured spans per model and tool call, tracking the four reliability signals — P95 latency, cost per run, tool error rate, and eval pass rate — so runs can be replayed, alerted, and scored. Start with one critical…

17 de ago. de 2026
sandbox
Analistas de segurança da informação

Bound an agent's blast radius with a deny-by-default sandbox policy — read allowed, writes whitelisted, egress denied — plus a per-server policy for each MCP server and violation telemetry. Contains a prompt-injected, trojaned, or rug-pulled agent; it does…

17 de ago. de 2026
build
Desenvolvedores de software

Orchestrate a spec-driven implementation pipeline: spec intake, requirements ledger, plan, architect gate, product gate, task decomposition with model tiering, parallel ralph-loop workers, a deterministic mechanical gate plus an opus judge, and verified…

16 de ago. de 2026
verdict
Analistas de garantia de qualidade de software e testadores

Zero-trust gatekeeper. Validates a deliverable claim-by-claim by re-running everything firsthand — refetch the diff from source, rerun the suite, re-execute the eval, rescore — never trusting the worker's say-so. Cycles work back when the letter passes but…

16 de ago. de 2026
capture
Analistas de garantia de qualidade de software e testadores

Turn a failure — a bug, a bad output, a prod incident, a verdict FAIL — into a minimal, reproducible eval level added permanently to the suite, so the same failure can never silently return. Builds the eval flywheel. Trigger: capture this failure, add a…

16 de ago. de 2026
chain-risk
Outras ocupações de informática

Compute compounding-error risk for a multi-step or multi-agent pipeline — per-step success rate raised to step count — and show where cumulative reliability falls below target, so you know where to place gates and checkpoints. Trigger: chain risk, compounding…

16 de ago. de 2026
constitution
Outras ocupações de informática

Pin a north-star constitution for a multi-agent system and protect the prompt layer from tampering — detect and block unauthorized system-prompt mutation and prompt-injection that tries to redirect downstream agents. Trigger: pin the constitution, north star,…

16 de ago. de 2026
Mostrando 8 de 19 skills coletadas.
Mostrando 1 de 1 repositórios
Todos os repositórios foram exibidos