Your colleague quit, but their skill didn't

What it does

dot-skill is a meta-skill that generates other skills for AI agent hosts. You feed it source material—Slack threads, WeChat exports, PDFs, or interview transcripts—plus a short description, and it emits a host-native skill that responds in the target’s voice and reasoning style. The result installs as a slash command inside Claude Code, Codex, DeepSeek Harness, Hermes, or OpenClaw.

The interesting bit

Instead of simple prompt injection, the project structures output into “Work” and “Persona” layers for colleagues, and ships a full six-dimension research toolchain for celebrities that tries to reconstruct decision frameworks rather than just catchphrases. It is essentially knowledge distillation applied to human relationships.

Key highlights

• Three distillation families: colleague (work standards + persona), relationship (personal, with photo-sharing planned), and celebrity (research-driven mental models).
• Native cross-host support: drops generated skills into each host’s standard skills directory, no wrapper required for DeepSeek Harness.
• Auto-collection from Feishu (API), DingTalk (browser automation), and Slack (Bot API, 90-day limit on free plans).
• Manual ingestion for WeChat SQLite exports, email .eml/.mbox, Markdown, PDFs, and images.
• Generated skills are versioned and can be rolled back with tools/version_manager.py.

Caveats

• DingTalk message history requires browser automation because the official API does not expose it.
• Slack auto-collection needs admin Bot installation and is limited to 90 days of history on free tiers.
• The relationship family’s photo-sharing feature is marked “coming soon,” so current skills are text-only.

Verdict

A pragmatic fit for teams that want to preserve institutional knowledge or experiment with persona-driven agents inside existing coding tools. Give it a pass if simulating departed colleagues—or ex-partners—inside your terminal feels more uncanny than useful.