Skip to content

Add Hypha to Agent Frameworks (tri-lingual) - #87

Merged
Zijian-Ni merged 2 commits into
mainfrom
add-hypha-framework
Aug 22, 2026
Merged

Zijian-Ni merged 2 commits into
mainfrom
add-hypha-framework

Conversation

@Zijian-Ni

@Zijian-Ni Zijian-Ni commented Aug 22, 2026 •

Copy link
Copy Markdown
Owner

Adds Hypha to Agent Frameworks in all three READMEs, following an outreach email from the CodeSoul team.

Why it belongs

Apache-2.0, TypeScript, public release v1.0.1 (2026-08-14), 15 aligned @codesoul-co/hypha-* packages on npm, 4 contributors, CI active. The design point is real and under-represented in this list: an Agent Core (ReAct, planning, tool selection, memory) split from a Production Harness (FSM execution, policy/approval, checkpoints, recovery, replay, audit), with product behaviour declared as versioned DomainPacks. Its "reuse without authority" rule β€” a cache hit may skip recomputation but can never authorise a tool call, advance the FSM, or replace event evidence β€” is a genuinely useful invariant.

Why the Unverified tag rather than a plain entry

It clears the 100+ star bar on paper (432 stars on 2026-08-22), but independent adoption has not shown up yet:

  1. Downloads are still small. @codesoul-co/hypha-core had 32 downloads in the last month as of 2026-08-22.
  2. The published benchmark is inside noise. The README reports tau-cubed 0.636 (245/385) for Hypha vs 0.626 (241/385) for a direct-model baseline β€” a 4-task, +1.04pp gap on single-trial runs. The 95% CI on that difference is roughly -5.8pp to +7.9pp. The entry describes the result without presenting it as a win. Credit to the vendor for publishing the direct baseline rather than omitting it.

The tag can come off once third-party adoption appears.

Verification

  • python3 scripts/sync_audit.py --summary -> RESULT: en/zh/ja fully in sync (858 entries each)
  • python3 scripts/check_markdown.py -> RESULT: markdown structurally clean
  • Repo facts pulled live from the GitHub and npm APIs on 2026-08-22

Not included

CodeSoul's Skill Garden is deliberately left out. The public repo is a usage guide; the actual Skill packages sit behind sign-in and eligibility checks, which fails the "publicly accessible" requirement in CONTRIBUTING.md.

Hypha (CodeSoul-co/Hypha) is an Apache-2.0 TypeScript agent framework
that separates an Agent Core from a Production Harness, with versioned
DomainPacks and a cache plane that cannot authorise side effects.

Listed with the Unverified tag rather than as a plain entry:

- The vendor's Skill Garden gates private Skill downloads behind
  starring this repo, so the star count is not a clean adoption signal.
- npm downloads for @codesoul-co/hypha-core were ~32/month on 2026-08-22.
- The vendor-published tau-cubed result (0.636 vs 0.626 over 385
  single-trial tasks) is inside statistical noise, so it is described
  as such rather than quoted as a win.

Verified 2026-08-22: v1.0.1 (2026-08-14), Apache-2.0, 15 npm packages,
4 contributors, CI active. sync_audit and check_markdown both pass
(858 entries, en/zh/ja fully in sync).
@github-actions

Copy link
Copy Markdown

πŸ›‚ PR Spam Guard report

Submitter: @Zijian-Ni β€” account age 1054 days, 8 followers, 29 public repos.
βœ… No automated red flags detected.
Recent PRs to awesome-* lists by this submitter (last 14 days):

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

πŸ’‘ Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: cad8305091

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with πŸ‘.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread README.md Outdated
- [Pydantic AI](https://github.com/pydantic/pydantic-ai) - πŸ†• ⚑ **v2.31.0 (August 15, 2026)** β€” near-daily release cadence; a 1.107.x maintenance line stays active. Agent framework from the Pydantic team that brings FastAPI-style ergonomics and real type safety to LLM apps β€” structured outputs validated by Pydantic models, dependency injection, streaming, and model-agnostic providers. The default pick when you want static typing and validated tool I/O rather than free-form prompt chaining. ![GitHub stars](https://img.shields.io/badge/dynamic/json?label=Stars&query=%24.stargazers_count&url=https%3A%2F%2Fapi.github.com%2Frepos%2Fpydantic%2Fpydantic-ai&color=yellow&logo=github&logoColor=white&style=flat&cacheSeconds=300)
- [Mastra](https://github.com/mastra-ai/mastra) - πŸ†• ⚑ TypeScript-first agent framework with workflow-driven development and built-in observability. **`@mastra/core@1.59.0` (August 14, 2026)**. ![GitHub stars](https://img.shields.io/badge/dynamic/json?label=Stars&query=%24.stargazers_count&url=https%3A%2F%2Fapi.github.com%2Frepos%2Fmastra-ai%2Fmastra&color=yellow&logo=github&logoColor=white&style=flat&cacheSeconds=300)
- [Agon](https://github.com/AutoResearch-Factory/Agon) - πŸ†• ⚠️ **Unverified** (35 stars, MIT). Autonomous omnidisciplinary research orchestrator built as a **Claude Code plugin** β€” scientist/coder/auditor multi-agent loop takes a bare topic all the way to running experiments with no human-written experimental code. 18 roles in 230.6 KiB of prompts across 10+ disciplines; 30-day continuous autonomous run documented. ![GitHub stars](https://img.shields.io/badge/dynamic/json?label=Stars&query=%24.stargazers_count&url=https%3A%2F%2Fapi.github.com%2Frepos%2FAutoResearch-Factory%2FAgon&color=yellow&logo=github&logoColor=white&style=flat&cacheSeconds=300)
- [Hypha](https://github.com/CodeSoul-co/Hypha) - πŸ†• ⚠️ **Unverified** (v1.0.1, August 14, 2026; Apache-2.0). TypeScript agent framework from CodeSoul that separates an **Agent Core** (ReAct, planning, tool selection, memory) from a **Production Harness** (FSM execution, policy/approval, checkpoints, recovery, replay, audit); product behaviour is declared as versioned **DomainPacks**, and a typed cache plane is explicitly barred from authorising side effects or advancing the FSM. 15 `@codesoul-co/hypha-*` packages on npm. ⚠️ Treat the star count with caution: the vendor's Skill Garden gates private Skill downloads behind starring this repo, npm downloads were ~32/month as of 2026-08-22, and the vendor-published τ³ result (0.636 vs 0.626 direct-model baseline over 385 single-trial tasks) is inside statistical noise. ![GitHub stars](https://img.shields.io/badge/dynamic/json?label=Stars&query=%24.stargazers_count&url=https%3A%2F%2Fapi.github.com%2Frepos%2FCodeSoul-co%2FHypha&color=yellow&logo=github&logoColor=white&style=flat&cacheSeconds=300)

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Link the non-obvious evidence directly

The new entry gives a release date, npm download count, Skill Garden gating claim, and benchmark results, but its only destinations are the repository root and a star badge. This prevents readers from verifying the claims and violates CONTRIBUTING.md:31-33, which requires dates and benchmark scores to link directly to primary sources; add direct release notes, npm statistics, gating-policy, and benchmark links to this entry and its two translations.

Useful? React with πŸ‘Β / πŸ‘Ž.

Keep the Unverified tag on adoption grounds only (npm downloads,
benchmark inside noise). The star-gating observation is removed from
the public entry.
@Zijian-Ni
Zijian-Ni merged commit 0e8fb5b into main Aug 22, 2026
2 checks passed
@Zijian-Ni
Zijian-Ni deleted the add-hypha-framework branch August 22, 2026 07:57
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant