repers/CITATION.cff at main · coolsocket/repers · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
cff-version: 1.2.0
title: "RePERS — operating layer for multi-agent repository work"
message: "If you reference this harness in academic work or technical writing, please cite it."
type: software
authors:
  - name: "coolsocket"
    website: "https://github.com/coolsocket"
repository-code: "https://github.com/coolsocket/repers"
url: "https://github.com/coolsocket/repers"
abstract: >-
  A contract layer for AI agents working on the same codebase at the same time,
  and for handing the resulting work off across teams, repos, and AI vendors.
  Where agent runtimes (LangGraph, CrewAI, OpenHands, Devin, etc.) execute LLM
  calls, RePERS sits above any runtime to coordinate them: preflight (capability
  discovery as shared memory across agents) → plan as a DAG with target-file
  isolation → parallel dispatch with collision guards → reviewer-on-join →
  evidence-based ship → portable release-pack handoff. A deterministic per-task
  router decides which slice of the pipeline a given task actually needs (one of
  skip / R-only / R-S / R-E-R / R-P-E-R / R-P-E-R-S), so the harness fires only
  when its shape matches the work. The value scales with codebase maturity:
  in a 200k-LOC monorepo with parallel multi-agent work the dispatch contract
  prevents N agents from clobbering each other; at enterprise scale the JSON
  evidence chain becomes an auditable handoff protocol between teams using
  different AI vendors; in cross-org settings the release-pack archive lets a
  receiving repo re-verify another team's work independently without trusting
  the sender's vendor. The dispatch contract is proven by a deterministic
  offline fixture before any live agent runs; every stage emits JSON evidence
  so a later agent (or human reviewer) can pick up where the previous one
  stopped. Ships a 25-entry capability registry, a Codex/Claude plugin with
  five skills (init / route / bug-hunt / release-pack / sinkin), a Git
  pre-commit hook, and four install fixtures proving cross-repo adoption
  from three different start states. Runtime is stdlib-Python only; the
  contracts are AI-agnostic by design.
keywords:
  - multi-agent
  - agent-orchestration
  - parallel-dispatch
  - operating-layer
  - agent-workflows
  - agentic-ai
  - ai-agents
  - claude-code
  - codex
  - dag
  - developer-tools
  - evidence-based
  - hooks
  - local-first
  - plugin
  - preflight
  - release-engineering
  - reproducible-builds
license: MIT
version: "0.1.0"
date-released: "2026-06-22"