s-kostyaev

Review code changes with a focused multi-agent workflow. Use when auditing a diff, branch, pull request, patch, commit range, or local worktree for bugs, regressions, missing tests, security issues, API breakage, maintainability risks, or review readiness.

Create or update Ellama Agent Skills. Use when designing new skills, porting skills from other agent systems, improving SKILL.md instructions, adding bundled resources, validating metadata, or adapting skills for small/local models.

Build, audit, or improve agentic coding/research systems and their harnesses. Use when designing agent workflows, small-model scaffolds, tool schemas, progressive-disclosure context, repository knowledge maps, guardrails, validation loops, subagents, evals,…

Efficiently capture web pages as clean markdown and query them with `mq` to avoid loading full content into context. Use when browsing web pages, extracting specific sections, or summarizing sources with minimal context usage.

Use this skill to browse the web textually.

Query markdown files efficiently with mq CLI. Use when exploring documentation structure, extracting specific sections, or reducing token usage when reading .md files.

Run DuckDuckGo web searches from the terminal using `ddgr` with JSON output. Use when the task calls for lightweight web search, quick result triage, or filtering by region/time/site.

Use this skill to generate changelog.

Use this skill when writing a concise commit message from a diff.

Conduct durable, source-grounded research with Sandwalk, including planning, parallel claims, immutable web or local-document snapshots, exact excerpts, reviewed findings, citation-safe drafting, recovery, and finalization. Use for research tasks that need…

Query Org-mode files and directories with oq using deterministic, structure-first, low-token workflows. Use when an agent needs to inspect .org content, locate sections/tasks/metadata, extract only scoped text.