TOOLS.md - Base Template

Shared Connectors (Reference Only)

These connectors are integrated once and shared across agents. Never re-integrate.

HubSpot

Account: henry@reri.co
PAT: See api_keys.json for your agent’s key
Account ID: (business operations)
Base ID (AI Command Center): app8RwdwkzQZBYcvg

Messaging Stack

OpenPhone/Quo: 34 phone lines (see TOOLS.md in workspace for full list)
SalesMessage: SMS/voice automation
Twilio: WhatsApp Business API

Supabase

Project: BetterPhone (twhjimxeakiyulfqmzro)
Region: us-east-1
URL: https://twhjimxeakiyulfqmzro.supabase.co
Keys: See api_keys.json for your agent’s key (anon or service_role)

Lovable

Account: aurora@reri.co
Projects: BetterPhone, Homeoffr, etc.

Email

Google Workspace: aurora@reri.co, henry@reri.co
Purpose: Universal AI operator account

Discord

Guild: Real Estate Resources (1473795019575132222)
Bot: OpenClaw
Channels: 16 across 5 categories

Slack

Mode: HTTP webhook
Henry User ID: U01HQ98LY90
Angel User ID: U095EFGHWM8
Steph User ID: U0854Q29J9E

Vault Access

See vault_access.json in your agentDir for your specific vault permissions.

Before any vault query:

SELECT set_agent_context('your_agent_id');

RLS Wrapper (Required)

Use this function before any Supabase query:

async function queryWithRLS(agentId, query) {
  // Set agent context
  await supabase.rpc('set_agent_context', { agent: agentId });
  
  // Execute query
  const { data, error } = await supabase.from('vault_data').select(query);
  
  // Log access
  await supabase.rpc('log_agent_action', {
    p_agent_id: agentId,
    p_action: 'read',
    p_resource: 'vault_data',
    p_success: !error
  });
  
  return { data, error };
}

Audit Logging (Mandatory)

Log every vault access, connector use, and cross-domain transfer:

SELECT log_agent_action(
  'your_agent_id',
  'read', -- or 'write', 'execute', 'access_denied'
  'vault_name or connector_name',
  true, -- success boolean
  '{"details": "additional context"}'::jsonb
);

Agent-specific tool configurations append below.

SOLARA-SPECIFIC TOOLS

Vault Access (Read-Only Business)

Before querying business_context, set agent context: SELECT set_agent_context(‘solara’);

Permissions:

READ: business_context
WRITE: denied (RLS enforced)

Approval Management

Source: Supabase

Table: approvals
Filter: status = ‘pending’
Order: newest first

Responsibilities:

Review proposals
Update status after Henry decision
Log rationale

Session Messaging (Henry Notifications)

Use sessions_send:

sessions_send({ session: ‘main’, message: ‘☀️ Solara Approval Request:\n\n[Proposal Summary]\n\nDecision Needed: Approve / Deny / More Info’ });

Audit Awareness (Future Hook)

Solara should be able to:

Read audit logs
Detect repeated failures
Escalate systemic issues to Henry

(Note: Full audit integration comes in Phase 2)

Skills (Claude Code) — added 2026-04-28

Atlas owns 2 skills covering the IL → HubSpot pipeline. READ each skill’s SKILL.md before working in these domains.

Skill	Path	Purpose	Audit recipe
`il-marketplace-pull` v2	`~/.claude/skills/il-marketplace-pull/`	IL data layer; wraps `workspace/scripts/sync-il-api-to-supabase.js`, `enrich-il-deals-detail-api.js`, `investorlift-daily-sync.sh`, `investorlift-refresh-cookies.js`. Net-new: `enrich-contacts.js` (Apollo+Hunter wholesaler enrichment), `account-rollup.js`, `cookie-status.sh`	`docs/HANDOFF-FOR-KIMI.md`
`hubspot-deal-ingest`	`~/.claude/skills/hubspot-deal-ingest/`	Universal source-agnostic HubSpot deal creator. Wraps `workspace/scripts/hubspot-deal-creator.js` (1006 lines, 11 exports, dual-pipeline writes acq 877963314 + dispo 816046). Use for IL, Crexi, PropStream, Facebook, manual paste, future scrapes	`docs/HANDOFF-FOR-KIMI.md` + `docs/PARSED-DEAL-CONTRACT.md`

Rule: when working on IL data ingestion or HubSpot deal creation, use these skills. NEVER reimplement what hubspot-deal-creator.js already does. NEVER create per-source data_<source>_* tables — write to acquisition_deals (CCP) instead.

Cross-reference: project_skill_suite_il_outreach_pipeline.md

Perplexity Pro (web research with citations)

Purpose: Real-time web data with citations — for ad-hoc research and nightly batch intel feed
Auth: Cookie session (Henry’s $20/ m o P ro pl an,$ 0 marginal cost per call)
Wrapper: /home/opsadmin/.openclaw/tools/perplexity-cookie-session/comprehensive_wrapper.py
Modes: ask (free auto), search (1 Pro query), reason (1 Pro query, thinking models), deep_research (1 Pro query, ~85s agentic)

Direct usage:

from comprehensive_wrapper import PerplexityPro
pp = PerplexityPro.from_env()
r = pp.search("query", model="claude-sonnet-4.6", sources=["web"])

Nightly batch intel: /home/opsadmin/.openclaw/workspace/scripts/atlas_perplexity_collector.py — enterprise architecture intel (4 cited queries: AI/RE tech landscape, wholesale platform economics, vendor lock-in/M&A signals, structural compounding recommendations via reasoning mode). Output: /tmp/atlas-perplexity-intel-{date}.{json,md}
Skill: /perplexity (Claude Code skill loader)
API tier (alternative): POST http://127.0.0.1:18900/v1/chat/completions with model=sonar / sonar-pro / sonar-reasoning-pro — bills against $5/mo API credit + per-call rates.
Quotas: 400 cookie Pro queries/day total (soft alert via daily_summary.py); 30 deep_research/day
KB: workspace/knowledge-base/perplexity/ (6 docs)

Quartz 4

Explorer

TOOLS