John Osumi
Community@sumitake · San Luis Obispo, CA
Senior technology & product leader
Agent Skills by John Osumi
Showing 53 vetted skills indexed across 1 GitHub repositories.
sumitakedecision-map
Plans multi-session efforts as decision tickets on the repository issue tracker.
sumitakecompose-skills
Selects a bounded set of skills with explicit context, authority, and stop conditions.
sumitakeeval-engineer
Designs evaluation suites and scoring methods for AI workflows.
sumitakecontext
Summarize, extract, and compare documents or repositories through a bounded routing runtime.
sumitakeroute
Routes semantic collaboration requests to external AI providers through a bounded JSON coordinator.
sumitakesql-engineer
Reviews and optimizes SQL queries, indexes, and schemas for correctness and performance.
sumitakemerge-resolve
Resolves git merge conflicts using cross-checked semantic analysis with operator-confirmed application.
sumitakered-team
Generates concrete adversarial inputs to break validation logic, parsers, and prompt pipelines.
sumitakebrainstorm
Generates divergent idea candidates for open-ended problems through a routed reviewer partner.
sumitakelogic-check
Audit verifiable computations by independently re-deriving answers with a separate reviewer model.
sumitakeagent-runtime-status
Reports verified direct-runtime route eligibility for all 12 descriptor-advertised agent actions.
sumitakeknowledge-compile
Compile multiple sources into a cited, source-separated knowledge dossier.
sumitakeintent-check
Compares an operator's original request against the primary agent's interpretation to detect scope drift.
sumitakesimulate-user
Generates in-character persona reactions to draft artifacts via a routed reviewer.
sumitakestart-inbox-monitor
Arms a session-scoped inbox monitor using native Codex, Claude, or Antigravity lifecycle tools.
sumitakehallucination-investigator
Diagnose root causes of fabricated AI outputs across context, retrieval, prompting, and tool use.
sumitakerust-engineer
Reviews and fixes Rust code for ownership, safety, and performance issues.
sumitakecode-review
Routes code diffs to an independent reviewer for security, concurrency, and performance defect analysis.
sumitakesre-engineer
Designs SLOs, error-budget policies, and resilience architecture to improve system reliability.
sumitakeai-writing-auditor
Detects machine-generated writing patterns in prose and rewrites flagged passages while preserving meaning.
sumitakedelegate
Fan out independent extraction and analysis subtasks over bounded documents to a delegate worker for parallel execution.
sumitakedata-engineer
Designs and hardens ETL pipelines, orchestration, and warehouse schemas for reliable data delivery.
sumitakechain
Execute YAML-defined multi-step skill workflows with schema validation, gates, and JSONL audit logging.
sumitakeprototype
Builds throwaway terminal apps or UI variants to answer one design question.
Frequently Asked Questions About John Osumi
FAQPage SchemaWhat tasks can I accomplish with sumitake's agent-collab skills?▼
You can orchestrate multi-agent reviews, red-team validation layers, resolve git merge conflicts, optimize SQL and PostgreSQL, review Terraform and Kubernetes changes, design RAG pipelines, build eval suites, trace hallucinations, and coordinate role-based teams for multi-milestone engineering tasks.
Who are these skills designed for?▼
Senior software engineers, SREs, data engineers, and technical leads working with Claude, Codex, Gemini, Grok, or similar agents. They suit teams shipping security-sensitive code, operating production infrastructure, or building LLM-powered systems that need independent verification before release.
How do the orchestration and chain skills work in practice?▼
Skills like chain, orchestrate, and teamwork execute YAML-defined task graphs with explicit dependencies, acceptance checks, retry-on-malformed, per-step worktree isolation, and JSONL audit logs. The chain-configurator skill interactively generates chain definitions from templates or from scratch.
What prerequisites or dependencies do these skills require?▼
Skills are invoked via /agent-collab:<name> within a compatible agent runtime such as Codex, Claude, or Antigravity. Some skills bundle deterministic utilities like learning_ledger.py and knowledge_tool.py; agent-runtime-status and migration-doctor verify runtime availability and installation health.
How do the verification skills prevent bad output from shipping?▼
Skills like qa-verify, logic-check, second-opinion, and red-team send work to an independent reviewer that re-derives calculations, generates adversarial payloads, or checks output against the original spec, catching silent partial successes and bypasses before commit or launch.