Detailed Analysis
A developer has released ProMa, an open-source Claude Code plugin designed to solve one of the most persistent friction points in AI-assisted software development: the loss of project context between sessions. Built as a suite of eight skills and one hook, ProMa maintains project state in plain markdown files, allowing Claude to resume work with full awareness of prior decisions, active tasks, and outstanding issues at the start of every new session. The tool is available on GitHub under an MIT license and includes a companion landing page, signaling an intent to build it into something beyond a personal utility.
The plugin's architecture reflects a sophisticated understanding of how agentic AI workflows break down in practice. ProMa introduces a two-queue task system that separates work Claude can execute autonomously from decisions that require human input, filing the latter into what the developer calls a "human inbox." This design addresses a well-documented failure mode in agentic systems: when an AI agent encounters an ambiguous decision point, it typically either halts, guesses, or hallucinates a resolution. ProMa's approach — having Claude file the decision as a pending item and move on to unblocked tasks — mirrors patterns from human project management methodologies and represents a practical application of human-in-the-loop principles at the workflow level rather than the model level. Additional features like architectural decision records (ADRs), staleness flags, daily sweeps with self-healing, and HTML or markdown dashboard output suggest the developer is modeling ProMa after enterprise project management conventions.
The broader significance of ProMa lies in what it reveals about the current state of Claude Code and agentic development tooling generally. Context persistence is a structural limitation of stateless large language model sessions, and the community is increasingly building infrastructure around this gap rather than waiting for it to be solved at the model or platform layer. That a developer found it necessary — and feasible — to construct a full project management layer in markdown as a plugin illustrates both the extensibility of Claude Code's plugin architecture and the maturity gap that still exists between what agentic AI can do in a single session versus what it can do across a sustained, multi-day project.
ProMa's design choices carry implications for how the broader AI development community may approach agentic workflow tooling going forward. By anchoring state in plain markdown and making everything git-friendly, the developer prioritizes auditability, portability, and composability over proprietary integration — a deliberate contrast to cloud-managed AI memory solutions. This positions ProMa within a growing ecosystem of community-built scaffolding that treats AI agents less as black-box assistants and more as components within traditional software engineering workflows, subject to version control, review, and structured handoffs. As Claude Code and similar agentic coding tools mature, developer-built plugins like ProMa are likely to inform the feature roadmaps of the platforms themselves, with solutions originating at the community layer eventually finding their way into first-party tooling.
Read original article →