Snipara OpenClaw Hooks

This package is not the same thing as snipara-companion. Use snipara-openclaw-hooks for OpenClaw-specific hook automation. Use snipara-companion for the general local helper CLI sourced from packages/cli.

Shared memory for multi-agent teams powered by Snipara.

Solo agent? OpenClaw's native memory system (MEMORY.md, session-memory) handles context well. These hooks add value when multiple agents share the same Snipara project — learnings from one agent benefit the whole team.

When to Use

| Scenario | Recommendation | | -------------------- | ----------------------------------------------------------------------- | | Solo agent | Use snipara-persist + snipara-stop only. OpenClaw handles the rest. | | Multi-agent team | Use all hooks. Shared context + team standards benefit everyone. | | Cross-project swarms | Full suite. Architectural decisions and learnings propagate. |

Hooks

Core Hooks (Solo or Team)

| Hook | Event | Description | | -------------------- | --------------------- | ------------------------------------------------------- | | 📝 snipara-persist | tool_result_persist | Auto-captures durable tool results with novelty gating | | 🛑 snipara-stop | command:stop | Saves deduped session context and commits durable notes |

Team Hooks (Multi-Agent)

| Hook | Event | Description | | ---------------------- | ----------------- | --------------------------------------------------------------- | | 🧠 snipara-startup | gateway:startup | Loads tiered session memory, tenant profile, and team standards | | 💾 snipara-session | command:new | Saves deduped context and reloads tiered memory | | 🚀 snipara-bootstrap | agent:bootstrap | Injects project docs and team coding standards |

Quick Start

# Solo agent — install only what you need
npx snipara-openclaw-hooks install snipara-persist snipara-stop

# Team/swarm — install everything
npx snipara-openclaw-hooks install

# Configure
export SNIPARA_API_KEY="rlm_your_key_here"
export SNIPARA_PROJECT_SLUG="your-project"

# Enable
openclaw hooks enable snipara-persist
openclaw hooks enable snipara-stop

Why Multi-Agent?

┌─────────────────────────────────────────────────────────────────┐
│  SINGLE AGENT                                                   │
│  └── OpenClaw native memory handles context fine                │
│  └── Snipara hooks = overhead, token waste                      │
│                                                                 │
│  MULTI-AGENT TEAM (Mike, Luna, Oscar...)                        │
│  └── Agent A learns: "API uses JWT, not sessions"               │
│  └── snipara-persist stores as learning                         │
│  └── Agent B queries: rlm_recall("authentication")              │
│  └── Gets Agent A's learning → no re-discovery                  │
│                                                                 │
│  VALUE = Shared learnings + Team standards + Less rework        │
└─────────────────────────────────────────────────────────────────┘

What Each Hook Does

snipara-persist (Recommended for all)

Auto-captures significant tool results with rlm_remember_if_novel so repeated test/build noise is skipped automatically:

| Pattern | Memory Type | Category | | ---------------------- | ----------- | ----------- | | git commit | decision | commits | | git push | fact | git | | pytest/vitest/jest | learning | testing | | build/compile/tsc | fact | builds | | deploy | decision | deployments | | error/fail/exception | learning | debugging |

snipara-stop

Extracts and saves deduped session context, then runs rlm_end_of_task_commit to keep only durable outcomes:

• Completed items
• In-progress work
• Blockers encountered

snipara-startup (Team only)

On gateway start:

• Loads rlm_session_memories (critical + daily tiers)
• Loads the latest tenant profile when present
• Injects team coding standards (if Team plan)

snipara-bootstrap (Team only)

Before workspace files load:

• Queries project documentation
• Loads team shared context (coding standards, best practices)
• Recalls project facts

snipara-session (Team only)

On /new command:

• Saves outgoing session summary with rlm_remember_if_novel
• Loads rlm_session_memories for the next session
• Loads the latest tenant profile when present

Automation Primitives

The hooks rely on the higher-level memory automation tools rather than only direct remember/recall calls:

• rlm_session_memories for session bootstrap with critical and daily tiers
• rlm_tenant_profile_get for client-specific constraints and preferences
• rlm_remember_if_novel to avoid storing duplicate tool output and session notes
• rlm_end_of_task_commit to preserve only durable decisions, learnings, and workflow changes

Canonical Event Alignment

The OpenClaw hooks package is aligning toward a canonical Snipara event model so multiple clients can feed the same ingestion pipeline.

Core canonical event types:

• session_start
• session_end
• compact
• message_user
• message_assistant
• tool_call
• tool_result
• file_changed
• error_observed

OpenClaw-specific hooks still emit native event categories, but hook payloads can now carry normalized fields such as:

• canonicalType
• sessionId
• workspace
• agentId
• privacyLevel

This keeps the adapter thin while preserving a shared backend contract across clients.

CLI Commands

# Install specific hooks
npx snipara-openclaw-hooks install snipara-persist snipara-stop

# Install all hooks
npx snipara-openclaw-hooks install

# List available hooks
npx snipara-openclaw-hooks list

# Check installation status
npx snipara-openclaw-hooks status

# Uninstall
npx snipara-openclaw-hooks uninstall --all

Configuration

Add to ~/.openclaw/openclaw.json:

{
  "env": {
    "SNIPARA_API_KEY": "rlm_your_key_here",
    "SNIPARA_PROJECT_SLUG": "your-project"
  }
}

Optional Settings

# Token budget for bootstrap context (default: 4000)
export SNIPARA_BOOTSTRAP_TOKENS=6000

# Custom API endpoint
export SNIPARA_BASE_URL="https://api.snipara.com"

Requirements

• OpenClaw installed (~/.openclaw/ directory exists)
• Node.js 18+
• Snipara account with API key

Get Your API Key

1. Visit snipara.com/dashboard
2. Create or select a project
3. Go to Settings → API Keys
4. Generate a new key (starts with rlm_)

Related

• Snipara Documentation
• create-snipara - Recommended base onboarding
• snipara-openclaw - External or legacy broader integration surface

Maintainer Release Notes

Changes under packages/snipara-openclaw-hooks/** are not auto-published from this repo.

Release flow:

1. validate the feature on dev
2. promote the validated commits onto a fresh branch from main
3. bump version in package.json
4. merge to main
5. run npm publish

If the same feature also touched backend logic or snipara-mcp, ship those surfaces together. See docs/operations/RELEASE_AND_DEPLOY.md.

License

MIT