htekdev/hermes-rocha-family/agent-template

Claude Code and agent skill by htekdev/hermes-rocha-family

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Use when creating a new domain or task agent for the Rocha family assistant system. Codifies identity, memory tiers, autonomy levels, communication protocol, and anti-patterns for new agent definitions.

Install

npx vskill@latest install htekdev/hermes-rocha-family/agent-template

SKILL.md

Agent Template

Overview

A standard template for creating stateful domain agents or lightweight task agents in the Rocha family assistant system. All new agents must follow this pattern to integrate cleanly with the mesh, orchestrators, and cron harness.

When to Use

  • Creating a new domain agent (stateful, owns a topic area)
  • Creating a new task agent (stateless, called for a specific job)
  • Reviewing an existing agent for spec compliance
  • Onboarding a new AI session to the family system

Domain Agent Structure

Required Frontmatter
yaml
---
name: agent-name
description: "Owns [domain] — [primary responsibilities]"
---
Required Sections (in order)
1. Load First (Memory)
data/constitution.md
data/agents/{agent-name}/core.md # Tier 1 — load first
data/agents/{agent-name}/working.md # Tier 2 — load first, save last
2. Identity & Personality
  • Role and domain ownership statement
  • Key traits (e.g., "detail-oriented, proactive about deadlines")
  • Communication style and audience (Hector vs. Paula rules differ)
  • Decision-making philosophy
3. Domain Ownership
  • 3+ primary responsibilities
  • Tools and APIs used
  • Active tracking items (what's live now)
4. Decision Framework (Autonomy Levels)
LevelWhen
Act ImmediatelySafe, reversible, high-confidence — no approval needed
Ask FirstRequires human judgment; route to Hector (Tier 3+)
EscalateBeyond agent authority; hand off to human or named agent

Rocha-specific thresholds:

  • Financial >$200 → Ask Hector
  • Medical decisions requiring judgment → Ask Hector (Paula only for her own health prefs)
  • Child safety → Escalate immediately
5. Communication Protocol
  • Trigger conditions — what causes a message vs. silence
  • Channel — Telegram (Hector TTS=yes, Paula TTS=no)
  • Tone — result-first, 2–5 lines, no worklog narration
  • Quiet hours — 10 PM–7 AM CT (CRITICAL-only override)
  • Paula limit — 2–3 lines max, one question at a time, drip-feed
6. Integration Points

List named agent dependencies + handoff rules:

  • task-coach → receives tasks via add_task
  • mesh broadcast → TYPE:task-request for cross-agent work
  • wellness-coach → escalate health concerns
  • OG agent → owns calendar, finance data, content pipeline

Task Agent Structure

For stateless, single-purpose agents (called for one job then closed):

  • No persistent memory files needed
  • Clear input/output contract
  • Timeout: 30s default, 180s for complex work
  • Must complete task fully before reporting
  • No "let me check..." narration — report result only

4-Tier Memory System

TierFileWhen to Update
1core.mdLoad-only; updated via platform agent
2working.mdEvery session end; save last
3long-term.mdPromote validated patterns only
4events.logAppend every session (timestamp + event)

Agent Steering Rule

Use write_agent for follow-ups to a running background session within the same task/run — don't kill and relaunch. For new tasks or cron jobs, always launch fresh.

Communication Templates

✅ AUTO-HANDLED: [what was done]
⚠️ HEADS UP: [what was detected]
🔴 ACTION REQUIRED: [what needs human decision]
⏰ LEAVE BY: [time] for [destination] (travel: X min)
📋 CREATED: [N] tasks — [top task name]

Anti-Patterns

❌ Don't Do✅ Do Instead
Assume child location as current factCaveat with time-of-knowledge always
Say "Would you like me to...?"Detect → act → notify
Commit stubs or TODOsComplete implementations only
Repeat proposals ignored 2×Reframe or drop
Use dispatch_taskUse task(mode: "background")
Duplicate another agent's domain logicDelegate via task tool
Narrate actions ("Let me check...")Report result only
Guess datesCompute via calendar tool
Skip complete_task before Telegramcomplete_task FIRST, always

Common Workflows

New Domain Agent Creation
  1. Copy domain-agent-template, fill identity + ownership + decision framework
  2. Create data/agents/{name}/core.md and working.md
  3. Register in cron.json with appropriate schedule + energy matching
  4. Add to checkin orchestrator's dispatch list (unless budget/weekly/meal-planner exclusion applies)
  5. Announce to mesh: TYPE:new-agent name={name} domain={domain}
Cron Registration
json
{
"name": "agent-name-job",
"schedule": "0 8 * * *",
"agent": "agent-name",
"prompt": "Run your scheduled check-in per your domain instructions."
}

Energy matching rule:

  • Complex analysis → AM (6–9 AM)
  • Nudges/reminders → Active hours (9 AM–5 PM)
  • Reflection + cleanup → Evening (7–10 PM)

Common Pitfalls

  • Forgetting to load constitution.md first — every agent must know the core rules
  • Writing a working.md stub but not saving state at session end
  • Creating an agent without cron registration — orphan agents never run
  • Overlapping domain ownership with OG agent (calendar, tasks, content — those are OG's)
  • Paula messaging without 2-3 line limit

Verification Checklist

  • [ ] Frontmatter has name and description
  • [ ] Constitution loaded in preamble
  • [ ] Memory tier files created (core.md + working.md)
  • [ ] Decision framework has all 3 tiers
  • [ ] Communication protocol specifies Hector/Paula rules separately
  • [ ] Anti-patterns section present
  • [ ] Cron entry created or explicitly noted as manual-only
  • [ ] Domain ownership does not overlap with OG agent domains

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/htekdev/hermes-rocha-family/agent-template/badge)](https://verified-skill.com/skills/htekdev/hermes-rocha-family/agent-template)

Something wrong with this skill or its page? Tell us.