Agent Shared Context

Cross-agent context sharing via shared files. Agents write trends, highlights, and signals to a shared folder. Other agents read before acting — creating coo...

Security Scan

The skill is internally consistent: it implements a simple file-based shared context (read/write JSON) exactly as described, with no external network, credentials, or surprising installs.

Purpose & Capability

Name/description match the provided code and instructions. The script reads/writes JSON in a shared workspace directory and the SKILL.md tells agents to call that CLI — all required capabilities are present and proportionate to cross-agent coordination.

Instruction Scope

Instructions direct agents to exec the included Python CLI to get/add/cleanup entries in shared JSON files — this is expected. Two notes: (1) the SKILL.md encourages embedding exec('python3 /path/to/scripts/...') in agent prompts which can introduce command-construction/injection risks if agent inputs are interpolated unsafely; (2) the script performs no file locking, so concurrent writers could race or corrupt state in multi-agent setups.

Install Mechanism

No install spec or remote downloads; the skill is instruction-only with a small included Python script. Nothing is written to disk beyond the shared JSON files the skill itself manages.

Credentials

No environment variables, credentials, or config paths are requested. The script uses only files under the workspace/shared directory, consistent with its stated purpose.

Persistence & Privilege

The skill is user-invocable and not always-enabled. It only creates/updates files within its own shared workspace; it does not modify other skills or system-wide settings.

Assessment

This skill is a simple, coherent file-based relay for agents, but it assumes a trust boundary: any agent or user with write access to the shared directory can influence all others. Before installing, ensure the shared folder is accessible only to trusted agents/users (set filesystem permissions), avoid constructing shell exec strings with unescaped/untrusted input (use safe argument passing), and consider adding file locks or atomic writes if you expect concurrent writes. If you need stronger guarantees, consider authenticated or signed entries or a small local service with access controls instead of plain JSON files.

Agent Relay — Cross-Agent Context Sharing

Lightweight file-based coordination for multi-agent OpenClaw setups.

Problem

Agents run independently — Reddit doesn't know what Moltbook found, News doesn't know what's trending on HN. Each agent is an island.

Solution

Shared JSON files that agents write to and read from. No config changes needed — just files on disk.

Setup

# Create shared directory in workspace
mkdir -p ~/.openclaw/workspace/shared

# Script location
~/.openclaw/workspace/scripts/shared-context.py

Usage

Agent writes after finishing work:

# Reddit agent logs trending topic
python3 scripts/shared-context.py add-trend \
  --source reddit --topic "self-hosting mini PCs" --score 1500

# News agent logs highlight
python3 scripts/shared-context.py add-highlight \
  --source news --title "Unsloth Studio launched" \
  --summary "Open-source LLM training UI, 2x faster"

Agent reads before acting:

# Moltbook agent checks what's trending before posting
python3 scripts/shared-context.py get-trends --limit 5

# Any agent checks recent highlights
python3 scripts/shared-context.py get-highlights --limit 5

Cleanup (brain-maintenance cron):

# Remove entries older than 48 hours
python3 scripts/shared-context.py cleanup --hours 48

Integration into cron prompts

Add to any agent's cron prompt:

Before posting/engaging, check shared context:
exec('python3 /path/to/scripts/shared-context.py get-trends --limit 3')
→ If a trend is relevant to this platform, create content about it.

After finishing, log what you found:
exec('python3 /path/to/scripts/shared-context.py add-trend --source <agent> --topic "<finding>" --score <relevance>')

Files

File	Purpose
shared/trends.json	Topics trending across platforms
shared/highlights.json	Best content found by any agent
scripts/shared-context.py	CLI to read/write shared context

Architecture

Reddit ──write──→ shared/trends.json ←──read── Moltbook
News   ──write──→ shared/highlights.json ←──read── Clawstr
                        ↑
              brain-maintenance (cleanup)

No database. No message queue. Just JSON files on disk. Works because all agents run on the same machine.