People's tacit judgment, communication norms and execution heuristics typically live scattered across chat logs, review comments and docs — and they vanish when someone leaves. dot-skill treats those traces as evidence and distills them into portable, inspectable Skills so an agent can make decisions and speak in the same frame of reference without relying on opaque hidden memories.
What Sets It Apart
- Two-layer Persona + Work pipeline: separates bounded behavioral voice (Persona) from procedural knowledge and output conventions (Work). This makes it easier to update tone independently of task heuristics, and to audit or roll back changes.
- Family-specific generation templates: supports three families (colleague / relationship / celebrity) with tailored collection and generation pipelines, so the system prioritizes different source types and consent constraints depending on use case.
- Multi-host, versioned deployment and correction lifecycle: generated Skills are versioned, accept natural-language corrections that write into a Correction layer, and can be installed or rolled back across supported agent hosts (Claude Code, Hermes, OpenClaw, Codex, DeepSeek Harness). This emphasizes auditability and iterative improvement rather than opaque prompt edits.
- Research tooling for public figures: a dedicated subtitles→transcript→merge→quality-check toolchain for celebrity-style skills supports reconstructing decision heuristics from long-form sources.
Who It's For and Tradeoffs
Great fit if you want to preserve and operationalize human expertise or communication style (e.g., departed teammates, mentors, creators) in an agent-friendly, inspectable package. It helps teams retain review standards, onboarding knowledge and reproducible communication norms. Look elsewhere if you lack high-quality source material (Skill quality scales with source quality), if legal/consent constraints make distillation inappropriate (relationship or celebrity variants have stricter assumptions), or if you need a production-grade system for safety-critical automation without additional governance: dot-skill packages human traces but does not by itself solve downstream legal/ethical risks.
How it works (brief)
The system ingests heterogeneous artifacts (chat exports, docs, emails, subtitles) and runs extraction and merge pipelines to populate two coordinated tracks: a capability/work track (procedures, heuristics, output templates) and a bounded behavior/persona track (tone, hard rules, correction history). Generated Skills are human-inspectable text packages that agents can invoke as installable modules.
