Context Bridge - No more re-explaining things on model change.

Maintains and transfers context across sessions, models, and time to avoid repeating information on every new interaction or model switch.

Security Scan

The skill's described purpose (cross-session context) matches most of its behavior, but there are inconsistencies and privacy risks (missing required tooling in metadata, repository/URL typos, and a setup script that scans and writes many local config files that may contain sensitive data).

Purpose & Capability

The SKILL.md and scripts implement registries, discovery, and a model-handoff memory which is coherent with the stated 'context bridge' purpose. However, metadata inside the packaged files (clawhub.json) declares 'jq' as required while the skill registry metadata at the top of the review said 'required binaries: none' — an inconsistency. There are also multiple different GitHub repository hostnames referenced (loudmouthedmedia, loudmoutclawmedia, openclaw), which suggests sloppy packaging or incorrect links.

Instruction Scope

Runtime instructions and the setup.sh script explicitly scan and read many user-local paths (~/.openclaw/workspace/skills, ~/.openclaw/skills, ~/.openclaw/agents, openclaw cron/agents output) and then create registry files under ~/.openclaw. Reading and aggregating SKILL.md, agent files, and cron payloads is functionally needed for a global registry, but it also means the skill will collect and write local configuration and any data contained there (which may include sensitive payloads or credentials). The SKILL.md also instructs maintainers to modify AGENTS.md to force session startup behavior, which changes agent startup semantics globally.

Install Mechanism

There is no formal install spec (instruction-only), lowering risk. But a provided scripts/setup.sh is intended to be run by users; the script depends on jq (and optionally the 'openclaw' CLI) even though top-level metadata omitted required binaries. The script will create and modify files under ~/.openclaw, so users must run it intentionally and inspect it before execution.

Credentials

The skill does not request environment variables or external credentials, which is appropriate. However, the setup process pulls data from local agent workspaces and cron payloads (via openclaw commands and file scans). Those sources can contain secrets or tokens; aggregating them into registries/discovery files may inadvertently centralize sensitive data. Because no explicit filters or redaction are visible, this is a proportionality/privacy concern.

Persistence & Privilege

The skill is not force-enabled (always: false) and is user-invocable. It recommends editing AGENTS.md so agents load the created registry files on startup — a legitimate integration point but one that changes agent startup behavior. The skill's ability to be invoked autonomously (disable-model-invocation: false) is the platform default and not flagged alone, but combined with broad file scanning it raises the blast radius if enabled without review.

What to consider before installing

Before installing or running setup.sh: (1) Inspect scripts/setup.sh yourself (don't run blind). The script uses jq and optionally the openclaw CLI — install jq or expect failures. (2) Verify which repository URL is authoritative (packaged files contain inconsistent GitHub links). (3) Backup ~/.openclaw and review current agent/crons/skill files — the setup will read and aggregate data from those locations and write registry files; those sources may contain secrets or cron payloads. (4) Consider running the script in a sandbox or on a copy of your ~/.openclaw tree first to see what it creates. (5) If you accept it, avoid blindly adding the recommended AGENTS.md startup lines until you confirm the registries contain only the information you want models to see. (6) If you have sensitive tokens in agent workspaces or cron payloads, remove or redact them before running the setup.

Context Bridge

The unified continuity system for OpenClaw.

Ensures context persists across sessions, models, and time. Prevents "starting from zero" on every /new.

Version: 1.0
Created: 2026-04-03

---

Quick Start

Installation

# 1. Install the skill
openclaw skills install https://github.com/loudmouthedmedia/context-bridge

# 2. Run setup to create registry files
cd ~/.openclaw/workspace/skills/context-bridge/scripts
./setup.sh

# 3. Update AGENTS.md (see below)

Setup Script

The setup.sh script auto-discovers and populates all existing skills, crons, and agents:

1. Scans ~/.openclaw/workspace/skills/ for local skills
2. Scans ~/.openclaw/skills/ for system skills
3. Checks openclaw cron list for active cron jobs
4. Checks openclaw agents list and ~/.openclaw/agents/ for agents
5. Creates and populates all registry files with actual data

---

What It Does

Before Context Bridge:

You: /new
Model A: "Hello! How can I help?" [blank slate]
You: [re-explain everything]

You: /new  
Model B: "Hello! How can I help?" [blank slate again]
You: [re-explain everything again]

After Context Bridge:

You: /new
Model C: "📧 [email-skill] available, 📈 [analytics-skill] ready. 
          Working on [Your Project]. Previous model 
          (GPT-X) scheduled a meeting 30 min ago. 
          Ready to continue."

---

Components

1. Registries (The "What Exists")

Registry	File	Tracks
Skills	~/.openclaw/skills-registry.json	All installed skills
Crons	~/.openclaw/cron-registry.json	Active scheduled jobs
Agents	~/.openclaw/agents/*/agent.md	Agent configurations
Discovery	~/.openclaw/skills-discovery.json	Skill capabilities

2. Handoff Memory (The "What Happened")

File: ~/.openclaw/model-agnostic-memory/model-handoff.md

• Session-to-session context
• What previous models did
• Active projects
• Recent actions

3. SOPs (The "Rules")

Documentation: ~/.openclaw/workspace/notes/openclaw-reliability-issues.md (Issue #9)

• File responsibility assignments
• Before-adding-content checklist
• Cross-reference comments

---

Session Startup Protocol

Required by AGENTS.md:

1. Read SOUL.md
2. Read USER.md  
3. Read memory files
4. READ CONTEXT BRIDGE FILES:
   - ~/.openclaw/skills-discovery.json
   - ~/.openclaw/model-agnostic-memory/model-handoff.md
   - ~/.openclaw/cron-registry.json
   - ~/.openclaw/skills-registry.json
5. ACKNOWLEDGE context in first response

---

File Responsibilities (SOP)

File	Responsibility	Must NOT
AGENTS.md	Session startup rules	Periodic checks
HEARTBEAT.md	Periodic health checks	Session setup
SOUL.md	Personality/vibe	Technical instructions
Context Bridge	Registries & discovery	Implementation

---

Usage

Automatic

On /new or model switch: Registries auto-load via AGENTS.md

Manual Fallback

If needed, say: load context

Update Workflow

When adding skills/crons/agents:

1. Update appropriate registry
2. Update discovery (if new capabilities)
3. Update handoff (log the change)
4. Git commit

---

Why "Context Bridge"?

• Context = What the model knows
• Bridge = Connection between isolated sessions/Models

Builds bridges across the memory gap.