All tools
SkillCommunity200

glebis Claude Skills

Updated Jul 7, 2026

A personal collection of Claude Code skills from Gleb Kalinin (glebis) focused on specialized, well-tested workflows. Skills include: a config-driven release manager with tiered compatibility policy and 27-unit-test stdlib engine; a synthetic coaching/therapy session generator across four therapeutic modalities; a semantic Obsidian vault search using the qmd engine with cross-lingual retrieval; evidence-based CBT/DBT intervention skills; and interactive HTML presentations with ElevenLabs narration.

View on GitHub

What it does

  • /release minor

    Cut a software release: bump version files, draft changelog from conventional commits, update COMPATIBILITY.md tiers, tag, and report closed issues.

  • /synthetic-session-generator

    Generate a persona-consistent synthetic coaching or therapy transcript in one of four therapeutic modalities (ICF/GROW, CBT, IFS, ACT/MI).

  • qmd-search (5 modes)

    Semantically search a local Obsidian vault using BM25, vector, hybrid, grep, or fused mode — with cross-lingual EN↔RU retrieval.

  • /present

    Generate an interactive HTML presentation with ElevenLabs voiceover narration synced to slides, optional GPT Image 2 illustrations, and Tufte typography.

Files (1)

OVERVIEW.mdprimary · markdown · 2.8 KB
## What it does

A personal collection of Claude Code skills by Gleb Kalinin (glebis), each built around a specific high-value workflow with a stdlib-only or minimal-dependency engine and bundled tests.

Skills in the collection:

Release (tiered compatibility) — config-driven release manager that bumps version files, runs a readiness gate, drafts a Keep-a-Changelog section from conventional commits, updates COMPATIBILITY.md with surfaces × maturity tiers (experimental/preview/stable), tags the release, and reports closed issues. A breaking change to a stable surface is forced to a major bump. Engine: 27-unit-test scripts/release.py; no third-party dependencies. Configure via release.config.json; invoke with /release minor or /release major.

Synthetic Session Generator — generate persona-consistent synthetic coaching and therapy session transcripts for evals, demos, and training data. Supports 4 modalities (ICF/GROW coaching, CBT, IFS parts-work, ACT/Motivational Interviewing), reusable persona bibles, 4 output formats (Fathom/Granola style, plain dialogue, structured JSON with eval labels, Obsidian markdown), 8 languages, and optional GPT Image 2 case-conceptualization cards. Always watermarked as synthetic.

qmd Search — semantic search over a local Obsidian vault using the on-device qmd engine (Node >= 22 or Bun). Five search modes: BM25 keyword, vector semantic, hybrid expansion+rerank, literal native-script grep, and fused (hybrid + grep). Cross-lingual retrieval (EN↔RU). Includes a smoke-test suite (18 cases) and quality evals (qmd bench).

Cognitive Toolkit (CBT/DBT) — evidence-based interventions: guided thought records, opposite action, DEAR MAN roleplay, crisis skills with HRV biofeedback. Configurable therapeutic pushback. Standalone or via Telegram.

Present — interactive HTML presentations with ElevenLabs voiceover narration synced to slides. Dual article/slides mode, scroll-reveal animations, optional GPT Image 2 illustrations, Tufte-inspired typography (EB Garamond + DM Sans).

GPT Image 2 — generate and edit images using OpenAI's GPT Image 2 model. 14 style presets, thinking mode (off/low/medium/high) for complex compositions, 8 platform presets.

## Install

    git clone https://github.com/glebis/claude-skills.git
    cp -r claude-skills/release ~/.claude/skills/
    cp -r claude-skills/synthetic-session-generator ~/.claude/skills/
    # copy whichever skills you need

For qmd-search, install qmd first:

    bun install -g @tobilu/qmd

## When to use

Use the release skill for any project that needs structured version bumping with tiered compatibility tracking. Use synthetic-session-generator for evaluation datasets or training data that requires domain-grounded fictional transcripts. Use qmd-search when you need semantic search across a local Obsidian vault without sending data to an external service.