aming-claw/docs/hn-demo/article.md at main · amingclawdev/aming-claw · GitHub
//blob/show" data-turbo-transient="true" />
Skip to content
Search or jump to...
Search code, repositories, users, issues, pull requests...
-->
Search
Clear
Search syntax tips
Provide feedback
--><br>We read every piece of feedback, and take your input very seriously.
Include my email address so I can be contacted
Cancel
Submit feedback
Saved searches
Use saved searches to filter your results more quickly
-->
Name
Query
To see all available qualifiers, see our documentation.
Cancel
Create saved search
Sign in
//blob/show;ref_cta:Sign up;ref_loc:header logged out"}"<br>Sign up
Appearance settings
Resetting focus
You signed in with another tab or window. Reload to refresh your session.<br>You signed out in another tab or window. Reload to refresh your session.<br>You switched accounts on another tab or window. Reload to refresh your session.
Dismiss alert
{{ message }}
amingclawdev
aming-claw
Public
Notifications<br>You must be signed in to change notification settings
Fork
Star<br>11
FilesExpand file tree
main
/article.md
Copy path
Blame<br>More file actions
Blame<br>More file actions
Latest commit
History<br>History<br>History
91 lines (74 loc) · 3.9 KB
main
/article.md
Top
File metadata and controls<br>Preview
Code
Blame
91 lines (74 loc) · 3.9 KB
Raw<br>Copy raw file<br>Download raw file
Edit and raw actions
Show HN: Aming Claw - A new multi-agent coding architecture (zero orchestration, commit-bound)
If you've used LangGraph supervisor, AutoGen GroupChat, or CrewAI for coding<br>work, you've written the orchestration. Aming Claw asks for zero.
The observer is your current Claude Code or Codex session, not a new daemon.
The observer holds the project's commit-bound code graph. It decides which<br>worker gets which files using two signals together: the requirement itself<br>(LLM-side) and the code graph's structural boundaries: dependency, module, and<br>function scope.
Each worker runs under its own contract: scoped files, fence token, trace<br>ledger, close gate. The full worker path runs each worker in an isolated git<br>worktree against a frozen commit hash. The HN demo starts from your current<br>Claude Code or Codex session as observer; scripted workers are a zero-setup<br>fallback that uses the same contracts, fences, and replay logic.
The shared object is not the chat. It is not the workflow state. It is the<br>project graph.
observer<br>commit-bound project graph<br>+------------+------------+<br>| |<br>Worker A contract Worker B contract<br>scope A, fence A scope B, fence B<br>| |<br>pass fail / interrupted<br>| |<br>candidate diff A replay B against X<br>| |<br>+------------+------------+<br>ordered Git merge<br>target graph reconcile once
The case I want you to challenge:
Worker A and Worker B both receive contracts bound to commit hash X.
Worker A passes; its diff is accepted as candidate evidence.
Worker B fails mid-execution.
The observer replays Worker B against commit hash X. Worker B sees the<br>original code, not Worker A's in-progress changes.
The replay passes, producing a clean diff against X — Worker B's contract<br>scope and Worker A's contract scope are disjoint by design, so B's replay<br>never touches files A already accepted.
Both accepted diffs land through an ordered Git merge.
The target project graph is reconciled once after the accepted change lands.
The backlog row closes only after the timeline and contract gates pass.
Worker A and Worker B can both be Claude, both Codex, scripted local workers, or<br>any compatible agent process. The coordination model is the same regardless of<br>runtime.
The installed-user demo starts with your current Claude Code or Codex session as<br>observer. Scripted workers are available for zero-setup reproducibility and CI,<br>so you do not need two AI subscriptions to challenge the protocol. Live worker<br>mode plugs in whichever AI runtime you have.
What is not new: supervisors, handoffs, traces, shared workflow state,<br>checkpoint replay, parallel branches. LangGraph has strong primitives for<br>supervisors, state graphs, checkpointing, replay, and durable workflows.
The narrow claim: I have not found another open-source, plug-and-play<br>coding-agent framework where:
the user writes zero orchestration code;
the observer decides scope from the project graph itself, not just the<br>prompt;
workers run under commit-bound contracts with fenced files and trace ledgers;
replay is tied to the original contract and frozen commit instead of chat<br>memory;
accepted work reconciles once against the target project graph before the<br>next agent treats it as truth.
If you know one -- research prototypes, commercial products, open-source<br>projects -- please send it to me. I'd genuinely like to know what to compare<br>against.
Repo: https://github.com/amingclawdev/aming-claw
How to run the demo: HN multi-agent challenge demo
More cases, audit trails, and the design story: Hope is not an engineering<br>control for AI coding agents
You can’t...