jkomoros/prompt-garden — Repo Appraisal

Overview

What it is — A TypeScript framework for composing LLM prompts as a directed graph of typed "seeds" that can be chained, remotely shared, and executed via CLI or webapp.
Problem — Resilient LLM prompt engineering requires breaking tasks into many cooperating sub-prompts; there was no standard way to version, compose, and reuse those prompts across projects.
Who it's for — Developers building chain-of-thought or tree-of-thought prompt pipelines who want a declarative, versionable, shareable prompt library format.
Notable — Declarative JSON seed-packets with JSON Schema autocompletion, remote packet loading by URL, associative memory backed by hnswlib, and a npm-installable runtime — an unusually complete approach for a 22-star repo, but development stalled in late 2023.

Verdict

	Rating	Summary
Quality	decent (14/24)	Solid documentation and engineering foundations undercut by total development inactivity since October 2023 and no tagged release.
PAI Relevance	watch (0.50)	The declarative seed-graph pattern is architecturally interesting but PAI's Fabric skill already covers structured prompt execution, and the repo is abandoned.

Quality Assessment

14/24 — stale-risk / adequately-documented / solid

Health: 3/8 (stale-risk)

Failed:

H1: FAIL — latest_release: "none" — no tagged release ever published
H2: FAIL — no release exists, so no recent release either
H3: FAIL — last commit 2023-10-21, over 2.5 years before appraisal date 2026-05-27
H4: FAIL — last commit 2023-10-21, well outside 30-day window
H8: FAIL — README contains no CI badge and no reference to .github/workflows

Passed:

H5: PASS — archived: false
H6: PASS — 56 open issues; above zero (active community at time of freeze) and below 100
H7: PASS — Apache-2.0 license declared

Documentation: 6/8 (adequately-documented)

Failed:

D5: FAIL — no heading matching API, Configuration, Options, Reference, Commands, or Parameters; sections are narrative ("Using the CLI", "Using the webapp") rather than reference-style
D7: FAIL — Discord invite link present but no docs site, wiki, or /docs directory link

Passed:

D1: PASS — README is present and extensive
D2: PASS — README is several kilobytes, well above the 1000-byte threshold
D3: PASS — explicit npm install, npm run build, and environment file copy instructions provided
D4: PASS — multiple fenced code blocks showing CLI invocation and Node package usage with output
D6: PASS — first paragraph states "A framework for gardening LLM prompts" and the Motivation section immediately follows
D8: PASS — prominent ⚠️ warning: "the format of local profiles… will likely change often, which means you might lose your state"

Engineering Signals: 5/8 (solid)

Failed:

E5: FAIL — 22 stars, below the 50-star threshold
E6: FAIL — 22 stars over ~35.3 months ≈ 0.62 stars/month, below 2/month
E7: FAIL — 3 forks, below the 5-fork threshold

Passed:

E1: PASS — primary language is TypeScript
E2: PASS — package.json present with full dependency manifest
E3: PASS — ~18 direct dependencies in dependencies block, below the 30-dep ceiling
E4: PASS — "test": "npm run test:base && npm run test:template" in scripts; mocha listed in devDependencies
E8: PASS — description "A framework for gardening LLM prompts" is meaningful and >20 characters

PAI Relevance

Dimension	Score	Assessment
Harvest Value	1	The typed declarative seed-packet format with JSON Schema validation and remote URL loading is a concrete design pattern worth studying for PAI's own prompt management; not novel enough to score 2 since the underlying chain-of-thought concept is standard.
Integration Readiness	1	TypeScript and available as `npm install prompt-garden`, so `bun add` is plausible; however the build chain uses the older `esm` shim and `mocha`, and the package is v0.0.4 with no releases — practical integration carries real compatibility and maintenance risk, warranting a moderate rather than full score.
Overlap Risk	1	PAI's Fabric skill already handles structured prompt pattern execution; prompt-garden's seed-graph model is distinct enough (declarative JSON, remote loading, associative memory) to avoid full overlap, but the functional territory partially overlaps with Fabric.
Gap Fill	1	PAI has no declarative prompt-graph composition system with remote packet sharing and typed schema validation; this partially fills that gap, though PAI's Fabric + existing agents cover the execution and chaining aspects at a practical level.

Composite: 0.50

What Next

Any multi-step LLM pipeline you're designing from scratch: Before committing to a workflow orchestration tool, sketch one pipeline using prompt-garden's seed-packet notation (typed JSON nodes, directed edges, schema-validated I/O) as a design exercise — not as a runtime dependency. The declarative format forces you to name inputs, outputs, and branching points explicitly, which surfaces ambiguities in the design before any code is written. The stalled development (late 2023) makes the library itself a poor runtime choice, but the schema pattern costs nothing to borrow as a planning artefact.
Prompt library fragmentation across CLI tools: The URL-based remote packet loading concept — load a shared prompt definition by HTTPS rather than vendor-copying JSON into each project — is worth watching independently of this repo's fate. Set a 6-month calendar bookmark to check whether this pattern has matured into any active TypeScript prompt tooling (LangGraph expressions, Pocket Flow, or a prompt-garden revival). If the pattern has landed somewhere maintained, it becomes a concrete candidate for centralising reusable prompt definitions across tools rather than duplicating them per-project.

Landscape Position

Category: LLM & Prompt Tooling

In this category: mattpocock--evalite (decent, 9/24 — previously scored weak; focused on evaluation rather than composition)

Standing: The only prompt-composition entry in the formal category; evalite targets evaluation, making these complementary rather than competing, but prompt-garden's abandonment in 2023 leaves the composition niche effectively unfilled by quality tooling.

Evidence Base

Density: 7/10 — README (extensive, ~8KB), dependency manifest (complete), repo metadata (stars, forks, dates, license, language) all available; missing: actual source file tree, test output, CI configuration, changelog, and any tagged release data.

Notes

The existing rolling-summary score of poor (1.6/24) for this repo appears to be a stale or quick-pass estimate; fresh probe-based scoring yields decent (14/24). The repo is conceptually ahead of its time — the seed-graph model with remote packet loading prefigures ideas now common in agent frameworks — but active development ended in October 2023 with 56 issues unresolved and no release ever cut. Worth revisiting only if the author resumes work or forks emerge.