Releases·Tool Release
Open-sourced BlaBla: executable project memory for coding agents
A developer released BlaBla, an MIT-licensed Rust CLI that moves project intent out of agent chat context into small executable contracts covering behavior, structure, and project layers. A verifier generates action sequences, tracks semantic coverage, and shrinks failures into reproducible counterexamples. In a four-stage handoff test with fresh Haiku sessions, BlaBla cost $0.62 versus $1.78 for growing prose while keeping architecture passing, though the author frames this as one observation on a synthetic project, not a general claim.
1 engagement1 sourceTue, Sep 15, 2026, 02:32 AM
AI Overview
- BlaBla is an MIT-licensed Rust CLI holding project intent in .bla contracts, with behavior, structure, and project layers verified to an OVERALL GREEN finish gate.
- A four-session Haiku test: BlaBla memory cut aggregate input tokens to 2.51M (vs 8.55M full prose) and cost to $0.62 (vs $1.78) with equal behavior scores.
- BlaBla's YELLOW state distinguishes unexercised obligations from GREEN; limits include heuristic verification, Python-only structure provider, and no temporal/distributed model.
Entities
Related
Products·launched
Yass – YAML spec syntax using RFC2119 keywords to guide LLM agents
Yass is a specification syntax that combines RFC2119 keywords with a minimal YAML format, leveraging LLMs' trained behaviors around those keywords to produce strong, consistent model outputs. A CLI is the only tool needed; agents read the repo README to proceed. It targets developers writing specs that LLM agents must implement reliably.
Yass is a specification syntax that combines RFC2119 keywords with a minimal YAML format, leveraging LLMs' trained behaviors around those keywords to produce strong, consistent model outputs. A CLI is the only tool needed; agents read the repo README to proceed. It targets developers writing specs that LLM agents must implement reliably.
1 engagement1 sourcehackernewsWed, Sep 23, 2026, 06:55 AM
Products·launched
Local-first AI agent harness for safe workspace automation
An open-source local-first agent harness that lets AI agents safely read and modify a workspace, run commands, and manage state. It includes approval workflows, subagent delegation, and long-term memory, aimed at developers and learners exploring AI agent internals.
An open-source local-first agent harness that lets AI agents safely read and modify a workspace, run commands, and manage state. It includes approval workflows, subagent delegation, and long-term memory, aimed at developers and learners exploring AI agent internals.
1 engagement1 sourceredditWed, Aug 19, 2026, 02:35 PM
Products·building
Portable agent state for switching between LLM harnesses
A system that standardizes and persists project context, coding conventions, workflows, skills, and memories across different agent harnesses, allowing users to switch agents mid-task without losing state. It solves the problem of hitting usage limits and losing context when switching between LLM coding agents.
A system that standardizes and persists project context, coding conventions, workflows, skills, and memories across different agent harnesses, allowing users to switch agents mid-task without losing state. It solves the problem of hitting usage limits and losing context when switching between LLM coding agents.
1 engagement1 sourceredditTue, Aug 25, 2026, 04:00 AM
Products·launched
ARGO: Open-source Agent Loop for traceable AI coding delivery
ARGO is an open-source framework that installs into coding environments like Cursor, GitHub Copilot, and OpenCode to make AI coding tasks traceable. It captures the chain from requirements to architecture to acceptance criteria, preventing loss of context across sessions and enabling better debugging by surfacing upstream assumptions. It solves the problem of opaque AI-generated code for developers and teams working on larger tasks.
ARGO is an open-source framework that installs into coding environments like Cursor, GitHub Copilot, and OpenCode to make AI coding tasks traceable. It captures the chain from requirements to architecture to acceptance criteria, preventing loss of context across sessions and enabling better debugging by surfacing upstream assumptions. It solves the problem of opaque AI-generated code for developers and teams working on larger tasks.
7 engagement1 sourceredditTue, Aug 4, 2026, 03:43 AM
Releases·Tool Release
Huuk: a rules-based steering plugin for Claude Code
A developer released Huuk, a plugin that lets Claude Code users define if-this-then-that rules in a readable file to gate actions like git push on tests, protect files, and auto-format after edits. It aims to give heavy agent-based developers more control over their coding agent.
A developer released Huuk, a plugin that lets Claude Code users define if-this-then-that rules in a readable file to gate actions like git push on tests, protect files, and auto-format after edits. It aims to give heavy agent-based developers more control over their coding agent.
2 engagement1 sourceredditWed, Jul 29, 2026, 09:33 AM