Prism Memory Starter

This starter is the smallest useful slice of the active prism-memory system. It gives you:

• the Python code for memory collection, digesting, rolling memory, knowledge validation, indexing, and the optional memory API
• the shell scripts that an agent can run for memory and knowledge operations
• starter config/state/inbox/knowledge scaffolding
• role docs and skill files you can load into an OpenClaw-like harness

It intentionally does not include live RaidGuild runtime data like raw transcripts, digests, rolling memory snapshots, or committed activity history.

What Is Included

scripts/                             agent entrypoints
skills/                              role skills + OpenClaw upload assets
docs/assistants/                     role boundaries and heartbeat docs
docs/collectors.md                   custom collector contract
examples/custom_collectors/          example fork-friendly collector scripts
superprism_poc/raidguild/code/       runnable Python packages
superprism_poc/raidguild/config/     starter space config
superprism_poc/raidguild/state/      starter state files
superprism_poc/raidguild/inbox/      shared intake lanes
superprism_poc/raidguild/knowledge/  starter evergreen KB scaffold

Quick Start

1. Copy .env.example to .env and fill in the collector credentials you actually use.
2. Review superprism_poc/raidguild/config/space.json and change bucket mappings, tags, and schedule times for your space.
3. Set PYTHONPATH=superprism_poc/raidguild/code.
4. Run the targeted commands from repo root:

python3 -m community_memory.pipeline collect --base superprism_poc --space raidguild
python3 -m community_memory.pipeline digest --base superprism_poc --space raidguild --date YYYY-MM-DD
python3 -m community_memory.pipeline memory --base superprism_poc --space raidguild --date YYYY-MM-DD
python3 -m community_memory.pipeline seeds --base superprism_poc --space raidguild --date YYYY-MM-DD
python3 -m community_knowledge validate --base superprism_poc --space raidguild
python3 -m community_knowledge index --base superprism_poc --space raidguild

Or use the wrappers:

bash scripts/memory_start.sh
bash scripts/knowledge_start.sh

Collector Model

Collector behavior is configured in superprism_poc/raidguild/config/space.json.

Built-in collector implementations shipped in the starter:

• discord_latest
• latest_meetings
• inbox_memory

Additional collector modes now supported:

• type: "python" for a collector class in a Python module
• type: "command" for a collector script or command declared in config

What a developer can change without code changes:

• enable or disable any built-in collector
• change window_minutes and initial_backfill_hours
• repoint env-backed endpoints like DISCORD_LATEST_URL and MEETINGS_LATEST_URL
• change bucket mappings and defaults in space.json

What still requires code changes:

• changing the payload contract expected by a built-in collector
• changing downstream digest or memory behavior

If you add an unknown builtin collector key to config.collectors, the pipeline skips it with a warning. If you want fork-specific collectors, use type: "python" or type: "command". See docs/collectors.md.

Required Environment

Collectors (configured in superprism_poc/raidguild/config/space.json):

Discord latest messages collector (discord_latest):

• DISCORD_LATEST_URL
• DISCORD_LATEST_KEY
• SPACE_HEAP_ID
• DISCORD_GUILD_ID

Latest meetings collector (latest_meetings, optional):

• MEETINGS_LATEST_URL for any HTTP endpoint that returns the payload shape expected by the built-in latest_meetings collector
• SPACE_HEAP_ID

Other optional environment:

• OPENCLAW_MEMORY_API_KEY for the API
• MEMORY_API_URL if you use scripts/memory_api_watchdog.sh

GitHub backup/push:

• GITHUB_OWNER
• GITHUB_REPO
• GITHUB_TOKEN

OpenClaw-Style Harness

There are two workable patterns.

1. Single automation agent

Use skills/openclaw/prism-memory-ops/ as the uploaded skill.

Recommended runtime assumptions:

• working directory: this repo root
• environment: values from .env
• command prefix: PYTHONPATH=superprism_poc/raidguild/code

Good first prompts:

• Run a memory pipeline health check, report collector state, and tell me what is missing before a full run.
• Validate the knowledge index state and rebuild indexes if needed.

2. Split specialist agents

For a harness that supports multiple agents or long-lived roles:

• memory role prompt source: docs/assistants/memory-manager/{IDENTITY,HEARTBEAT}.md
• knowledge role prompt source: docs/assistants/knowledge-manager/{IDENTITY,HEARTBEAT}.md
• memory role skill: skills/memory-manager-subagent/
• knowledge role skill: skills/knowledge-manager-subagent/

Suggested commands:

• memory role: bash scripts/memory_start.sh
• knowledge role: bash scripts/knowledge_start.sh
• pause a lane: bash scripts/agent_pause.sh memory|knowledge|all
• resume and sync a lane: bash scripts/agent_unpause_sync.sh memory|knowledge

Notes

• The starter keeps the existing raidguild path and --space raidguild defaults so the code works immediately. If you want a different space slug, update config, commands, and any harness prompts together.
• scripts/agent_unpause_sync.sh, scripts/memory_start.sh, and scripts/knowledge_start.sh assume the repo has a Git remote and can pull/push. Remove or edit those steps if your harness runs against a local-only checkout.
• The API docs are in docs/README.md.
• Collector docs live in superprism_poc/raidguild/code/community_memory/pipeline.py, superprism_poc/raidguild/code/community_memory/collector.py, and superprism_poc/raidguild/config/space.json. There is not a separate plugin-authoring guide yet.
• Custom collector authoring is documented in docs/collectors.md.