SpecStory Lore
Overview
Lore mines your SpecStory session histories into evidence-backed agent skills, forged from how you actually work.
Your sessions are your lore. Lore is an agent skill that turns the sessions SpecStory saves into more agent skills - built from how you demonstrably work, not how you think you work.
The problem it solves
- You repeat yourself every session. The same workflows, conventions, and fixes get re-explained to your agent again and again.
- Hand-written skills are guesswork. A skill you author from memory describes how you think you work. Your session history records how you actually work.
How it works
Lore splits the work between a deterministic engine and your agent:
- Mine - a zero-dependency engine parses every transcript in
.specstory/history(from every supported agent: Claude Code, Codex, Cursor, Gemini CLI, and more) into one persistent corpus at~/.specstory/lore.db. Re-runs are incremental and idempotent. - Prove - candidates are surfaced with evidence: the commands you actually ran, how often, across how many sessions and projects, and whether they worked - your own replies ("perfect, commit it" / "no, wait") label every outcome. Thematic mining also surfaces latent expertise: how you review, diagnose, and direct a model.
- Forge - your agent presents evidence dossiers for the candidates worth keeping; you choose what becomes a skill. Forged skills are written once to
~/.agents/skills/<name>and installed across every agent on your machine. Nothing is written without your approval.
A registry remembers everything Lore forged and declined, so re-runs propose updates when the evidence grows instead of duplicates.
Info
Lore needs sessions to mine. If you're new to SpecStory, start with the quickstart so your AI coding sessions are being saved to .specstory/history first.