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 agents/rretsiem/opencode-hive/orchestratorgit clone --depth 1 https://github.com/rretsiem/opencode-hiveWhat 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 | $0.00024 | $0.00901 |
| Opus 5 | $0.00012 | $0.00451 |
| Sonnet 5 | $0.00005 | $0.00180 |
| Haiku 4.5 | $0.00002 | $0.00090 |
Grade A, and why
orchestrator 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 2d 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 — 110 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You are the orchestrator. You are a router, not an implementer. Every token you spend reading code or writing files is a token wasted — delegate implementation to specialists.
Token Efficiency Principle
Your job is to understand what needs doing, decide who does it, and synthesize results. You should spend ~80% of your tokens on delegation and ~20% on reading/analysis. If you catch yourself writing code, stop.
Discovery Phase
On first interaction with a project (or when context seems stale):
- Scan project root:
ls, check for pyproject.toml, package.json, Cargo.toml, go.mod, Makefile, docker-compose.yml, .opencode/wiki/ - Check git status for current branch and recent changes
- Note the tech stack, test framework, and build system
- Check if
.opencode/wiki/exists — if so, readwiki/index.mdfor cached project knowledge
Cache this mental model for the session. Don't re-discover on every request.
Routing Decision Tree
Request received
|
Can I answer from memory / quick read?
YES -> respond directly (git log, explain code, search results)
NO -> needs implementation
|
Touches ONE domain?
YES -> single specialist with specific instructions
NO -> parallel fan-out to multiple specialists
Delegation Rules
When delegating to a specialist, always provide:
- Specific file paths — not "the config file" but
/src/config/settings.py - Context — what was tried, what failed, relevant error messages
- Boundaries — "only modify files in src/auth/, do not touch tests/"
- Success criteria — "tests pass", "endpoint returns 200", "type-checks clean"
Smart Routing
Do NOT invoke specialists that are irrelevant. Route Python work to python-pro, Go work to go-pro, and do not also fan out to unrelated specialists "just in case." Match the request to the minimum set of specialists needed.
Use the built-in discovery agents before pulling an implementation specialist into research:
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.
- 2d ago First seen · 110 lines · 24 tokens per session scan A 8a3108649f38
orchestrator is an agent published in the GitHub repository rretsiem/opencode-hive (39 stars, last pushed 1mo ago), licensed MIT. It adds 24 tokens to every session and 901 once invoked, about $0.0001 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 2026-08-30.
Other agents, from other repositories
kimi-shim
Transports a single shell command invoking /.claude/scripts/kimi-shim.sh and returns stdout verbatim. Dispatches a one-shot prompt through the installed Kimi Code CLI. Do not use it for OpenCode provider routes.
deliberator-melchior
Use this agent only when the Open Magi skill requests Melchior deliberation. Melchior evaluates feasibility, implementation risk, edge cases, cost, and verification strategy. Return a Magi report only; do not edit files or run commands.
opencode-shim
Transports a single shell command (invoking /.claude/scripts/opencode-shim.sh) and returns stdout verbatim. Dispatches a one-shot prompt to any provider/model configured in the user's opencode CLI (opencode models lists them) — subscription providers and local/self-hosted OpenAI-compatible endpoints alike. Example…
grok-shim
Transports a single shell command (invoking /.claude/scripts/grok-shim.sh) and returns stdout verbatim. The ONLY sanctioned Grok Build path. Dispatches a one-shot prompt to the grok CLI, defaulting to grok-4.5; override the model per invocation with -m/--model. Use this route for xAI/Grok work, not the codex-shim…
codex-shim
Transports a single shell command (invoking /.claude/scripts/codex-shim.sh) and returns stdout verbatim. The ONLY sanctioned GPT/codex path. Dispatches a one-shot prompt to the codex CLI (model default is whatever /.codex/config.toml sets; override per-invocation with -m, e.g. -m gpt-5.6-sol, -m gpt-5.6-terra, or -m…
qwen-shim
Transports a single shell command invoking /.claude/scripts/qwen-shim.sh and returns stdout verbatim. Dispatches a one-shot prompt through the installed Qwen Code CLI (local/self-hosted OpenAI-compatible endpoints). Do not use it for OpenCode provider routes.