Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/humanerd-drew/opencode-drewgent/agent-profilesnpx skills add humanerd-drew/opencode-drewgent --skill agent-profilesgit clone --depth 1 https://github.com/humanerd-drew/opencode-drewgentWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00062 | $0.01690 |
| Opus 5 | $0.00031 | $0.00845 |
| Sonnet 5 | $0.00012 | $0.00338 |
| Haiku 4.5 | $0.00006 | $0.00169 |
Grade A, and why
agent-profiles scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 3d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 150 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agent Profiles
Agent profiles are pre-defined subagent roles loaded by task(subagent_type="<name>", description="summary", prompt="..."). Each profile sets model, provider, toolsets, and system instructions in one call.
Location
Profiles live at $HERMES_HOME/agents/*.md. For {{AGENT_NAME}}: ~/.{{AGENT_NAME_LOWER}}/agents/.
Format
Each profile is a Markdown file with YAML frontmatter and markdown body:
---
name: profile-name
description: >
Brief description of this profile's purpose.
model: deepseek-v4-flash
provider: opencode-go
toolsets: [terminal, file, search]
created: YYYY-MM-DD
---
# Profile Name
You are a [role definition]. Your job is to...
## Rules
- Specific constraints (read-only, write-only, review-only, etc.)
## Escalation
When creating new autonomous capabilities, **prefer assigning a single agent profile over building a multi-stage pipeline.**
| Approach | When | Why |
|----------|------|-----|
| **Profile-first** | Agent acts autonomously (observe → decide → act) | Simpler, faster to iterate, one agent owns the full cycle |
| **Pipeline-first** | Multi-stage processing with distinct roles | Necessary when stages need different models/tools or parallel execution |
**Origin:** The CMO agent (content-manager) started as a 5-stage pipeline design. The user corrected it: "워크플로우 만들어두고 에이전트 배정하고 기록 뒤져서 바로 만들어볼 수 있지 않겠니." The simpler approach won — one agent profile + cron scheduling + one narrative tracking file. Pipelines add accidental complexity. Start with a profile; split into stages only when there's a clear reason.
## The 8 Standard Profiles
### Flash Tier ($0 marginal cost — OpenCode Go subscription)
| Profile | Model | Role | ESCALATE? |
|---------|-------|------|-----------|
| **explorer** | deepseek-v4-flash | Read-only codebase analysis, context gathering | ✅ |
| **implementer** | deepseek-v4-flash | Code implementation, file creation | ✅ |
| **tester** | deepseek-v4-flash | Test writing + verification | ✅ |
| **archiver** | deepseek-v4-flash | Documentation, changelog, completion summary | ❌ |
### Pro/Max Tier (stronger models, selective use)\n\n| Profile | Model | Role |\n|---------|-------|------|\n| **content-manager** | deepseek-v4-pro | CMO agent — observes work, produces multi-format content (blog + SVG cover + X thread + Excalidraw PNG), tracks narrative arc. Runs daily via cron, delivers to Discord. |\n| **reviewer** | deepseek-v4-pro | General code review (logic, style, edge cases) |\n| **reviewer-critical** | qwen3.7-max | In-depth review for large/architectural changes |\n| **security-reviewer** | qwen3.7-max | Security audit (auth, crypto, injection, secrets) |\n| **planner** | qwen3.7-max | Task decomposition, tier assignment, pipeline design |
## Pipeline Pattern
Sequential multi-stage workflows use individual `task()` calls:
```python
task(
subagent_type="explorer",
description="Analyze auth code",
prompt="analyze the existing auth code"
)
task(
subagent_type="implementer",
description="Implement login validation",
prompt="implement login validation"
)
task(
subagent_type="tester",
description="Write tests",
prompt="write tests for login"
)
task(
subagent_type="reviewer",
description="Review auth changes",
prompt="review the auth changes"
)
Or via kanban pipeline automation:
kanban_create(
title="Add login validation",
pipeline=["explorer", "implementer", "tester", "reviewer", "archiver"],
body="...",
)
Tiered Pipeline by Complexity
Tier 1 (trivial): Implementer(flash) → Archiver(flash) — typo fix, rename
Tier 2 (moderate): Explorer(flash) → Implementer(flash) ↔ Tester(flash) [≤2 loops] → Archiver(flash)
Tier 3 (complex): Planner(max) → Explorer(flash) → Implementer(flash) ↔ Tester(flash) [≤3 loops] → Reviewer(pro) → [security?] → Security-reviewer(max) → [critical?] → Reviewer-critical(max) → Archiver(flash)
Design Principles
1. Never assume provider topology. The user may route through any provider (OpenCode Go, OpenRouter, MiniMax direct, etc.). Always verify the actual config before designing routing strategy.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 3d ago First seen · 150 lines · 0 tokens per session scan A 633fd1cab3bf
agent-profiles is a skill published in the GitHub repository humanerd-drew/opencode-drewgent (2 stars, last pushed 1mo ago), licensed MIT. It adds 62 tokens to every session and 1,690 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
agent-comm-hub
本地多智能体通信 Hub(MCP stdio / HTTP-SSE),提供消息、任务编排、共享记忆、进化引擎,暴露 58 个 MCP 工具 + Web 管理面板.
Elephant Agent
Complete guide to what Elephant Agent is, how to use it from the CLI, how herd, wake, Personal Model, proactive curiosity, skills, and tools fit together, and how to explain Elephant Agent clearly to users, contributors, or evaluators.
Claude Code
Guides operator-owned delegation to Claude Code when repo work benefits from a second coding lane or an interactive code agent.
Codex
Guides operator-owned delegation to Codex when the task fits a bounded coding lane, review pass, or worktree-isolated implementation track.
OpenCode
Frames OpenCode as an operator-controlled coding lane for bounded implementation or review work inside a real repository.
skill-hub-gateway-setup
Configure Skill Hub Gateway (OpenClaw) - install skill, obtain API key via bootstrap flow, and set up environment variables.