Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions CLAUDE.md
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,7 @@ No build step -- this is a pure documentation repo. Browse by topic:
- **Vector database (Weaviate):** `Knowledge/README.md`
- **Prediction models:** `Predictions/README.md`
- **Language models (8B):** `Generation/README.md`
- **Agent infrastructure:** `Infrastructure/README.md`
- **n8n workflow automations:** `Automation/README.md`

## Architecture
Expand All @@ -17,6 +18,7 @@ No build step -- this is a pure documentation repo. Browse by topic:
Knowledge/ Weaviate vector-graph database docs (connection, search, RAG, schema)
Predictions/ HuggingFace text regression models (performance + preference prediction)
Generation/ 8B language models (Llama 3.1 base, continual pre-training + instruct)
Infrastructure/ Clean-room agent runtime, orchestration, and tool-safety patterns
Automation/ n8n workflow templates for advocacy automation
.github/ Dependabot config + CI workflows
```
Expand All @@ -28,6 +30,7 @@ Automation/ n8n workflow templates for advocacy automation
| `Knowledge/README.md` | Weaviate connection details, search ops, RAG patterns, Content schema |
| `Predictions/README.md` | Prediction model usage, batch processing, score clipping |
| `Generation/README.md` | 8B model usage, generation parameters, known limitations |
| `Infrastructure/README.md` | Clean-room agent runtime roadmap across scanner, platform, and tooling repos |
| `Automation/README.md` | n8n hosting options, workflow import, activation |
| `.gitleaksignore` | Secret scanning exclusions (read-only API keys in docs) |

Expand Down
97 changes: 97 additions & 0 deletions Infrastructure/README.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,97 @@
# Clean-Room Agent Infrastructure

This guide describes how Open Paws can adopt stronger agent-runtime patterns without copying proprietary implementations. The goal is to reuse the ideas that are broadly useful, then implement them from scratch in our own systems, data models, and interfaces.

## Core Principles

1. Build from patterns, not copied code
2. Keep sensitive workflows explicit and reviewable
3. Centralize tool definitions, permissions, and metadata
4. Separate long-running orchestration from user-facing chat
5. Preserve source provenance for scanner findings, quests, and automations

## Shared Runtime Pattern

The target shape across Open Paws projects is:

1. **Scanner and external signal ingestion**
2. **Structured export and normalization**
3. **Platform-side task, quest, and conversation orchestration**
4. **Tool registry with sensitivity-aware UI and routing**
5. **Human review and approval before high-impact actions**

This keeps research, execution, and user interfaces aligned while still letting each repo own a clear slice of the system.

## Current Implementation Tracks

### `Open-Paws/project-compassionate-code`

The scanner is the best place to normalize repository findings into a stable export format. That export should include:

- deterministic finding identifiers
- repository provenance
- merge-tier or effort metadata
- stable payloads that downstream systems can ingest without re-parsing scanner output

This repo should continue to act as the source of truth for machine-readable contribution opportunities.

### `Open-Paws/open-paws-platform`

The platform should own guild-facing orchestration and persistent state. Current and near-term responsibilities include:

- ingesting scanner findings into draft guild quests
- storing quest source provenance
- supporting dry-run previews before import
- acting as the eventual home for cleaner plan, task, and background-session orchestration

This is the right layer for approval boundaries, operator controls, and durable workflow state.

### `LarytheLord/Open-Paws-Tools-Platform`

The tools platform is the right sandbox for experimenting with more agentic UX before it is promoted into production systems. The first clean-room pattern to establish here is a central tool registry that can drive:
Comment on lines +47 to +51
Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

🧹 Nitpick | 🔵 Trivial

Consider varying word choice for better readability.

The word "right" appears twice in close proximity (lines 47 and 51: "right layer" and "right sandbox"). While not incorrect, using a synonym in one instance would make the text flow more smoothly.

✨ Proposed alternative wording
-The tools platform is the right sandbox for experimenting with more agentic UX before it is promoted into production systems. The first clean-room pattern to establish here is a central tool registry that can drive:
+The tools platform is the appropriate sandbox for experimenting with more agentic UX before it is promoted into production systems. The first clean-room pattern to establish here is a central tool registry that can drive:
📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change
This is the right layer for approval boundaries, operator controls, and durable workflow state.
### `LarytheLord/Open-Paws-Tools-Platform`
The tools platform is the right sandbox for experimenting with more agentic UX before it is promoted into production systems. The first clean-room pattern to establish here is a central tool registry that can drive:
This is the right layer for approval boundaries, operator controls, and durable workflow state.
### `LarytheLord/Open-Paws-Tools-Platform`
The tools platform is the appropriate sandbox for experimenting with more agentic UX before it is promoted into production systems. The first clean-room pattern to establish here is a central tool registry that can drive:
🧰 Tools
🪛 LanguageTool

[style] ~51-~51: You have already used ‘right’ in nearby sentences. Consider using an alternative word to let your writing stand out and sound more polished.
Context: ...ls-Platform` The tools platform is the right sandbox for experimenting with more age...

(REP_RIGHT)

🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.

In `@Infrastructure/README.md` around lines 47 - 51, The repeated use of "right"
reduces readability; update one of the phrases—either change "right layer" or
"right sandbox"—to a synonym (e.g., "appropriate," "ideal," or "proper") in the
Infrastructure/README.md section that includes "right layer for approval
boundaries, operator controls, and durable workflow state" and the
"LarytheLord/Open-Paws-Tools-Platform" paragraph to improve flow while keeping
meaning intact.


- quick actions
- intent detection
- sensitivity labels
- future approval prompts
- future policy hooks for restricted workflows

This keeps the UI and runtime metadata from drifting apart as more tools are added.

## Recommended Next Steps

### Layer 1: Shared Definitions

- standardize tool metadata fields across repos
- align sensitivity levels for public, medium-risk, and high-risk workflows
- document a shared scanner export contract

### Layer 2: Orchestration

- add explicit background-task lifecycle states
- separate draft generation from action execution
- add review checkpoints before external writes or sensitive research flows

### Layer 3: Operator Experience

- show sensitivity cues directly in the UI
- make provenance visible for imported quests and generated tasks
- support dry-run previews wherever automation can create records or tasks

## Safety Boundaries

Open Paws should not copy source files, prompts, comments, or proprietary internal naming from third-party leaked codebases. Safe reuse means:

- extracting product patterns
- rewriting implementations from scratch
- documenting trust boundaries
- validating that sensitive workflows have clear operator review paths

## Success Criteria

This infrastructure direction is working when:

1. scanner findings move into the guild pipeline with stable provenance
2. tooling interfaces share one source of truth for intent and risk metadata
3. higher-sensitivity actions are visible, reviewable, and easier to gate
4. repo-specific experimentation can graduate into the platform without rewriting the architecture each time
5 changes: 5 additions & 0 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,11 @@ Documentation for our specialized generative AI models:

- [Overview & Quick Start](Generation/README.md)

### 🧠 Agent Infrastructure
Documentation for clean-room agent runtime, orchestration, and tool-safety patterns across Open Paws projects:

- [Overview & Quick Start](Infrastructure/README.md)

### 🤖 n8n Workflow Automations
Documentation and example workflows using n8n to automate advocacy tasks.

Expand Down
Loading