Skill Synthesis

import { Card, CardGrid } from ‘@astrojs/starlight/components’;

Skill Synthesis

The third time you do the same thing, Ptah notices.

Skill Synthesis watches your sessions for trajectories — sequences of (turn count, tool calls, outcome) that look like reusable workflows. When the same trajectory succeeds 3 times without matching an existing skill (cosine ≥ 0.85 on the embedding), Ptah materialises it as a permanent skill at:

~/.ptah/skills/<slug>/SKILL.md

From that point on, any agent can invoke it like a hand-authored skill — same trigger semantics, same on-the-fly context injection.

What’s in this section

Trajectory extraction, dedup, promotion. [Learn more →](/skill-synthesis/how-it-works/) Promote, reject, or just watch what Ptah found. [Learn more →](/skill-synthesis/reviewing-candidates/) What the generated file looks like. [Learn more →](/skill-synthesis/skill-md-anatomy/) Thresholds and caps. [Learn more →](/skill-synthesis/settings/)

Why it exists

Hand-authored skills are great when you know up-front what’s worth abstracting. Most workflows aren’t like that — they emerge from repetition. Skill Synthesis catches those without you having to notice them yourself.

Limits

Active skills are capped at 50 by default (LRU eviction — the least-recently-used skill is archived when the cap is hit). Cosine dedup against the active set prevents near-duplicates from polluting the directory.