agentsclimarketplace

Br3eze code

Skill br3eze-code/br3eze-code

Personal Ai AgentFrom the repository description

Install
npx -y skills add br3eze-code/br3eze-code

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

11.9 KB, ~3.2k tokens by cl100k_base, as published. Nobody here has run it

AgentOS Skill Reference — v2026.7.0

AgentOS uses a modular 3-tier ReAct reasoning engine. The AI (Gemini 2.0 Flash / Claude Sonnet 4.6 / GPT-4o / Ollama) automatically discovers, chains, and executes these skills via the Tool Registry and Skill Loader.


Architecture: 3-Tier Ask Engine

Every user query is routed through three escalating tiers:

Tier 1: Keyword match → direct tool execution (0 LLM calls, <5ms)
         e.g. "active users", "system stats", "who", "dhcp leases"

Tier 2: Regex rule → action function (0 LLM calls, <5ms)
         e.g. "kick alice", "ping 8.8.8.8", "block 192.168.1.5", "gen voucher 1Day"

Tier 3: LLM ReAct loop (1–5 LLM calls, 500ms–5s)
         System context = memory + skill + lang hint + EmotionEngine tone hint
         Max 5 tool-call turns before forced return
         Dispatches to: manage_network / manage_vouchers /
           manage_finance / manage_mesh / manage_memory / manage_project

Tool Registry (32 Built-in Tools)

Permission Tiers

TierTools
read-onlysystem.stats, users.active, users.all, dhcp.leases, arp.table, interfaces, wireless.clients, ip.routes, voucher.stats, voucher.list, finance.report, finance.trends, finance.audit
workspace-write (default)user.add, user.remove, user.kick, firewall.block, firewall.unblock, voucher.create, ping, traceroute, dns.flush
danger-full-accesssystem.reboot, system.backup, wireless.set_frequency

All tools flow through PermissionPolicy and HookRegistry (onBefore / onAfter hooks for audit logging, SSE broadcast, and gateway activity push).


Skill Groups

1. MikroTik Hotspot Management (manage_network)

Full WiFi hotspot lifecycle on MikroTik RouterOS v7.

User Management:

  • user.add — Create hotspot user (profile, password, limit)
  • user.remove — Delete user
  • user.kick — Disconnect active session
  • users.active — List currently connected users
  • users.all — List all provisioned users

Network Inspection:

  • dhcp.leases — Show DHCP leases
  • arp.table — Show ARP table
  • interfaces — List interfaces and status
  • wireless.clients — List wireless associations
  • ip.routes — Routing table
  • ping <host> — Ping test
  • traceroute <host> — Traceroute
  • dns.flush — Flush DNS cache

Firewall:

  • firewall.block <ip|mac> — Add drop rule
  • firewall.unblock <ip|mac> — Remove drop rule

System:

  • system.stats — CPU, RAM, uptime, interfaces
  • system.reboot — Safe router reboot (danger-full-access)
  • system.backup — Create RouterOS backup (danger-full-access)
  • wireless.set_frequency — Change wireless channel (danger-full-access)

Security Rules:

  • Never expose passwords in responses
  • Always confirm destructive actions (kick / remove / reboot)
  • All changes logged to rolling audit buffer (last 1,000 events)
  • executeCLI() rejects shell metacharacters and enforces 4096-char limit

2. Voucher & Billing System (manage_vouchers)

Plans:

PlanDurationPrice (USD)
1hour1 hour$1.00
1Day24 hours$5.00
7Day7 days$25.00
30Day30 days$80.00

Code format: STAR-[A-F0-9]{6} (e.g., STAR-4A2F8C)

Tools:

  • voucher.create <plan> — Create & provision voucher on MikroTik; generates QR code; optionally initiates Mastercard A2A payment
  • voucher.list — Show active/recent vouchers (limit, used filters)
  • voucher.stats — Usage and revenue statistics
  • voucher.redeem <code> <user> — Redeem voucher for user

Recurring Billing (Auto-Renewal Engine):

  • guardHotspot reaper monitors session expiry every 1 hour
  • If user has hasPlan status and sufficient wallet balance → auto-deduct credits and extend MikroTik session without interruption
  • Falls back to expiry + notification if wallet insufficient

Lifecycle:

createVoucher() → provisioned on MikroTik → QR generated →
customer redeems → session expires →
Auto-Renewal Engine (checks wallet) → renews or kicks →
expireOldVouchers() marks used

3. Financial Engine (manage_finance)

Wallet & Revenue:

  • finance.report — Revenue report (all-time summary)
  • finance.trends — 7-day revenue trend data
  • finance.audit — Audit trail (last N events)

Enhanced P2P Credit Transfers: Users can transfer credits using identifiers beyond UIDs:

  • Resolvers: Phone (E.164), Email, or Username
  • Fee Logic: Configurable via P2P_FEE_PERCENT and P2P_FEE_FLAT env vars
  • Dual-Entry: Separate p2p_transfer_sent and p2p_transfer_received records

Mastercard A2A Payments:

  • voucher.payment.initiate — Initiate Account-to-Account payment for voucher purchase
  • voucher.payment.status <paymentId> — Check payment status
  • Webhook-based reconciliation via POST /api/webhook/mastercard
  • OAuth 1.0a RSA-SHA256 signing

4. Project Tracking: CPM + EVM (manage_project)

Critical Path Method (CPM):

  • Forward pass: ES, EF computed via recursive dependency traversal
  • Backward pass: LS, LF, float
  • Critical path: all tasks where float === 0

Earned Value Management (EVM):

MetricFormula
SPIEV / PV
CPIEV / AC
SVEV − PV
CVEV − AC
ETC(BAC − EV) / CPI
EACAC + ETC
VACBAC − EAC
TCPI(BAC − EV) / (BAC − AC)

EVM data sources: voucher revenue as EV/PV; LLM token spend + infra cost as AC.

Project Tools (9 actions):

  • project.create — Create project with tasks and dependencies
  • project.update — Update task progress / actuals
  • project.critical_path — Compute and return critical path
  • project.evm — Calculate EVM metrics
  • project.report — Full CPM + EVM summary report
  • project.list — List all projects
  • project.get <id> — Get project details
  • project.delete <id> — Delete project
  • project.export — Export project data

5. Agent Memory (manage_memory)

Vector + KV memory over Firebase / local backend:

  • memory.store <text> — Save a memory entry (with embedding)
  • memory.get — Retrieve all memory entries
  • memory.forget <key> — Delete a memory entry
  • Cosine similarity search for relevant memory retrieval in LLM context

6. Multi-Router Mesh (manage_mesh)

  • nodes.list — List all registered mesh nodes
  • nodes.register — Register a new router node
  • nodes.exec <node> <tool> — Execute tool on a specific node
  • mesh.exec <tool> — Execute tool across all nodes simultaneously

7. Messaging Channels

AgentOS supports four unified messaging channels with consistent command dispatch and admin authorization.

Supported Channels:

  • Telegram — Bot API polling, ALLOWED_CHAT_IDS allowlist, /claim setup wizard
  • WhatsApp — Baileys socket, JID allowlist, E.164 normalized
  • Slack — Workspace integration, /dashboard, /stats commands
  • Discord — Guild integration, matching command parity

Public Customer Flow:

Inbound message
  → detectLanguage() [en|es|fr|sw] + classifyIntent() [BUY|FIX|BALANCE|ADMIN|IDLE]
  → EmotionEngine.update()
  → BUY    → sendButtons() with plan selection
  → FIX    → AskEngine.run() → diagnostic reply
  → BALANCE → balance check
  → IDLE   → AskEngine.run() → AI general reply

Admin Commands (all channels):

/setup    → multi-step wizard: IP → user → pass → _finishSetup()
/status   → system + router overview
/voucher  → voucher management
/users    → active users list
/stats    → revenue dashboard
/dashboard → operator dashboard
/doctor   → health check for all 4 channels + Firebase + MikroTik

Security: Admin command debounce 3s, rate limit 30 req/60s per chatId. All channels share BaseChannel.isAuthorized().

Cross-Channel Email Identity:

  • Messaging handlers extract email from inbound messages
  • Firebase Auth verifies email identity and prevents duplicate records
  • Channel-specific IDs (Telegram chatId, WhatsApp JID) mapped to centralized Firebase UIDs
  • Authenticated email is the authoritative identity source

8. EmotionEngine

Per-chat state machine (capped at 10,000 entries):

State: { mood: [-1,1], urgency: [0,1], energy: [0.2,1], trust: [-1,1] }

Signals:
  FIX intent     → urgency += 0.4, energy += 0.15
  BUY intent     → mood += 0.05
  Negative words → mood -= 0.15
  Positive words → mood += 0.12, trust += 0.05
  Per message    → energy -= 0.05 (floor 0.2)
  After TTL      → urgency -= 0.3

Tone hints injected into LLM system prompt:
  urgency > 0.6  → "Be concise and direct. Prioritise resolution."
  mood < -0.3    → "Be empathetic and solution-focused."
  trust > 0.7    → "Good rapport. You can be friendly."

9. Self-Audit Engine

Synchronous scan of rolling audit buffer (last 1,000 events):

Anomaly TypeConditionSeverity
rate_anomalySame actor > 10 sensitive actions in 1 hourHigh
instant_redeemVoucher created and redeemed within 5s by same actorMedium
  • Scheduled every 6 hours via AgentOSOrchestrator._scheduleSelfAudit()
  • On-demand via GET /api/audit/self
  • Violations broadcast to all admin chats and WS gateway

10. Configuration & Diagnostics

  • config.get <path> / config.set <path> <value> — Read/write config
  • config.show — Display full CONFIG object
  • doctor — Run health check across all channels (Telegram, WhatsApp, Slack, Discord), Firebase, MikroTik; auto-fix where possible
  • status — Quick system + router overview (CPU, RAM, uptime, active sessions)
  • system.monitor — CPU/RAM threshold alerts every 15s

CLI Commands (REPL + WebSocket)

help        status      active      users       kick <u>
voucher <p> vouchers    redeem <c> <u>
ping <h>    logs [n]    dhcp        arp
firewall    block <t>   unblock <t> reboot
adduser     deluser     connect     disconnect
agent <q>   nodes       memory      tools
cli <cmd>   api <path>  audit       clear       exit

Cron Schedule (AgentOSOrchestrator)

JobScheduleAction
daily_reboot04:00Router maintenance reboot
heartbeat12:00Admin broadcast: system alive
Voucher expiryEvery 1hMark and notify expired vouchers
Self-auditEvery 6hScan event log for anomalies
New device detectionEvery 1mAlert on unknown MAC addresses
System monitorEvery 15sCPU/RAM threshold alerts

HTTP API Quick Reference

Public:

  • GET /health — System health, voucher stats, metrics snapshot
  • POST /voucher/redeem — Redeem voucher (code, user)
  • GET /voucher/:code/qr — QR code PNG
  • POST /api/webhook/mastercard — Mastercard A2A webhook

Authenticated (Authorization: Bearer <GATEWAY_TOKEN>):

  • GET /api/stats — Router + DB stats
  • GET /api/vouchers — List vouchers
  • POST /tool/execute — Execute any tool by name
  • GET /api/ask/stream — SSE streaming AI query
  • POST /api/ask — Single AI query
  • GET /api/audit/self — On-demand self-audit
  • GET /api/finance/summary — Revenue summary
  • POST /api/voucher/payment/initiate — Mastercard A2A initiation
  • GET /api/nodes / POST /api/nodes — Mesh node management

Adding New Skills

New skills can be added by creating files in skills/ and registering them in the SkillRegistry. The Skill Loader discovers SKILL.md trigger patterns at startup and matches them during Tier-2 routing.


AgentOS v2026.7.0 — Built for Africa's last-mile operators. Br3eze Africa · github.com/br3ezeafrica

What ships with it: 1333 files

9955.2 KB alongside SKILL.md, 580 of them executable

adapters/

1293 more files not listed here. See all 1333 in the repository.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.