A practice framework for thinking with AI
| Rating | Summary | |
|---|---|---|
| Quality | decent (13/24) | Actively maintained with a thoughtful README, but no dependency manifest, no tests, and low adoption signal keep the engineering score at floor. |
| PAI Relevance | WATCH (0.38) | The ritual initialization pattern and workshop-as-memory framing offer study value, but Python language and no callable CLI make direct integration impractical. |
13/24 — maintained / adequately-documented / no-signals
Failed:
Passed:
Failed:
Passed:
Failed:
Passed:
| Dimension | Score | Assessment |
|---|---|---|
| Harvest Value | 1 | The "summoning ritual" pattern — structured system prompts that initialize an agent into a defined cognitive stance — and the workshop-as-memory design offer study value for PAI's skill context architecture, but do not introduce a fundamentally novel pattern PAI hasn't approached through its existing skill/hook system. |
| Integration Readiness | 0 | Python-tagged with no dependency manifest or described CLI; the framework is primarily content and conventions rather than callable code, requiring significant adaptation to extract patterns into PAI's TypeScript/Bun skill system. |
| Overlap Risk | 1 | Partial overlap with PAI's existing file-based memory system (WORK/, LEARNING/, KNOWLEDGE/) and the Telos/Interview skills that handle goal-tracking and conversational context fill — but the explicit practice-ritual and stance-initialization layer is not directly replicated. |
| Gap Fill | 1 | PAI has no dedicated structured practice-session framework with a defined AI stance and explicit capture/process/orient cycle; Telos and Interview cover adjacent territory but not the same ritual cadence and session-arc pattern. |
Composite: 0.38
If you maintain a recurring research or writing practice with AI: Run one week of sessions using only Magic's core ritual — open a plain markdown workshop.md, write a brief orientation note at the top of each session, and append decisions and open questions before closing. At the end of the week, review whether the accumulated context measurably changed the quality of the next session's starting point. This tests the workshop-as-memory pattern in practice before committing to the full framework structure.
If you are designing any agent initialization workflow: Note the pattern of using static markdown files as a context inheritance layer — the agent reads prior state from a file rather than relying on platform memory or session history. This is model-agnostic and editor-agnostic. Bookmark the repo's prompt and ritual structure for reference when you next need a portable, reproducible way to warm-start an agent with accumulated context, without depending on any vendor's memory feature.
Revisit in 4–6 months: The repo's core design idea is coherent but the implementation is early — no versioning discipline, limited community signal, and the framework depends heavily on individual practice consistency rather than enforced structure. Check back when it has documented examples of sustained multi-week practice runs or community-contributed workshop templates, which would validate whether the ritual layer actually produces compounding value at scale.
Category: Personal AI & Knowledge
In this category: tinyhumansai--openhuman (excellent, 21/24); malteristo--magic is the first practice-framework entry in this subcategory.
Standing: Magic occupies a distinct practice-oriented niche within personal AI — lower quality signal than openhuman but conceptually differentiated by its ritual structure, explicit AI stance design, and portability-first philosophy.
Density: 9/10 — README (full, 8KB), repository metadata (stars, forks, dates, language, license, release), topic tags, and landscape context all available; dependency manifest absent and source code not directly inspected.
Magic is the framework layer above malteristo--turtleos (its companion self-hosted infrastructure repo, also in the vault). The two repos form a deliberate pair: Magic defines the cognitive patterns and rituals; turtleOS makes them persistent and local-first. Any future evaluation of either should consider both together. The rolling summary previously assigned poor (2.6/24) — that figure appears to have been a preliminary pass; probe-based scoring produces decent (13/24).