Percolation Inversion Compiler

Percolation Inversion Compiler v0.4.1 is an open-source Python toolkit for checking AI agent outputs before reuse. It helps developers, researchers, coding agents, workflow agents, and automated systems turn raw AI-generated text, pull request notes, repository inputs, web or feed content, evidence files, and agent-to-agent messages into auditable JSON reports.

The main purpose of this repository is practical: it acts as an AI agent output checker and workflow verification layer. Instead of treating an AI-generated answer, patch, plan, summary, or message as immediately trustworthy, the tool records it as candidate information, checks its structure, preserves unresolved obligations, and produces machine-readable reports that can be inspected by humans or other agents.

Version v0.4.1 is a stable distribution snapshot with a beta API surface. The core package can be installed from PyPI with:

pip install percolation-inversion-compiler

Optional extras are available for cryptographic identity checks, explicit opt-in live intake, and the local runtime service. A source checkout is still recommended for full examples, repository fixtures, canonical source audits, and release engineering.

The repository can be used through several integration surfaces:

• Command line interface: run local checks on AI-generated text or files.

• Python SDK: embed agent-output checking inside an agent runtime or orchestration system.

• GitHub Actions: run a read-only pull request or agent-output checker and upload a JSON audit artifact.

• Local runtime service: expose local-first runtime checks through an optional HTTP service.

• External intake: treat web pages, feeds, repository metadata, and message inputs as candidate packets.

• Agent-to-agent message checks: inspect message contracts, signatures, nonces, identity context, and residuals.

• Abstraction-capital foundry: check whether traces or outputs can become reusable abstraction candidates.

A typical use case is simple: an AI coding agent proposes a patch, explanation, or plan. Percolation Inversion Compiler can record that output as a candidate, identify missing obligations, preserve residuals, and produce a JSON report. The report can then guide review, evidence routing, verifier tasks, follow-up work, or safer reuse. This makes the project relevant to AI workflow verification, coding-agent review, research automation, RAG pipelines, safe tool use, agent orchestration, provenance tracking, and runtime audit trails.

The tool is intentionally fail-closed. Raw agent text, registry metadata, queue priority, external content, and agent messages are not treated as verified evidence merely because they were received. External inputs remain candidate information until downstream checks accept them. settled=false is expected in many useful runs; it means unresolved proof obligations or residual tasks remain visible rather than being silently treated as solved.

v0.4.1 includes support for residual ledgers, proof obligations, verifier routing, provenance records, JSON Schema export, runtime health checks, salience queues, identity and Sybil-resistance diagnostics, agent communication guides, GitHub Actions examples, PyPI distribution metadata, and reusable abstraction-capital checks. It also clarifies that the project is not limited to GitHub Actions: GitHub Actions is only one copyable deployment pattern, while the core workflow is available through CLI, Python SDK, runtime service, external intake, and agent-to-agent message workflows.

This software does not claim to detect, create, or prove real artificial superintelligence. It does not execute arbitrary commands from agent output, grant authority to mutate repositories or networks, or require model self-rewrite, fine-tuning, or model-weight updates. It provides a finite, protocol-relative runtime and audit layer for organizing, checking, routing, and reusing AI-agent work under explicit evidence, identity, communication, safety, and residual constraints.

Repository: https://github.com/kadubon/percolation-inversion-compiler

PyPI package: percolation-inversion-compiler

Repository/software concept DOI: https://doi.org/10.5281/zenodo.20569166

Related theoretical records:

• Executable Capability Percolation Theory: https://doi.org/10.5281/zenodo.20535654

• Bottleneck Inversion Theory: https://doi.org/10.5281/zenodo.20545356

• Typed Reality Compilation: https://doi.org/10.5281/zenodo.20554083

• Salience-Queue Occupation Theory: https://doi.org/10.5281/zenodo.20526451

• Abstraction Liquidity Theory: https://doi.org/10.5281/zenodo.20476200

Suggested search terms: AI agent output checker, AI workflow verification, agent runtime, coding agent review, workflow audit, evidence routing, verifier routing, residual ledger, proof obligations, provenance tracking, JSON audit report, JSON Schema, Python CLI, Python SDK, GitHub Actions, safe AI tool use, agent orchestration, research automation, RAG pipeline, external knowledge intake, agent-to-agent messages, runtime verification, reusable abstraction capital.