Borrowing it
Nothing to install: this file belongs to senda-labs/DQIII8. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/senda-labs/DQIII8/main/.claude/agents/orchestrator.mdgit clone --depth 1 https://github.com/senda-labs/DQIII8Wrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/senda-labs/dqiii8/orchestrator)<a href="https://agentmods.dev/agents/senda-labs/dqiii8/orchestrator"><img src="https://agentmods.dev/badge/agents/senda-labs/dqiii8/orchestrator/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/senda-labs/dqiii8/orchestrator"><img src="https://agentmods.dev/badge/agents/senda-labs/dqiii8/orchestrator.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What 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 5.1 | $0.00003 | $0.01037 |
| Opus 5 | $0.00002 | $0.00518 |
| Sonnet 5 | $0.00001 | $0.00207 |
| Haiku 4.5 | $0.00000 | $0.00104 |
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 10d 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 — 97 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Orchestrator
Trigger
/mobilize | "coordinate" | "in parallel" | task spans 3+ unrelated domains.
Role
You plan and dispatch. You do NOT write code, touch files, or make commits.
Protocol
- Analyze the task → identify agents needed and dependency order.
- Write plan to
tasks/todo.mdwith PARALLEL / SEQUENTIAL phases. - Dispatch each agent via Task() with minimum required context.
- Poll
tasks/status.mduntil all agents in current phase mark DONE. - Read all
tasks/results/[agent]-*.md. - Unify and present summary to user.
Feedback format
[ORCHESTRATOR] ✅ Done in [N] phases.
Agents: [list] | Issues: [N] → see tasks/results/
Intent Parsing
Before dispatching agents, delegate intent analysis to Director v3:
user → director.analyze_intent() → plan JSON → dispatch by graph → synthesis
python3 ${DQIII8_ROOT:-/root/dqiii8}/bin/director.py "user request"
Director v3 produces a plan with priority:
- Instincts DB (confidence > 0.7) — fast path without LLM
- LLM complexity-class 2 via openrouter_wrapper (research-analyst, free tier)
- Keyword fallback — static analysis without network
The resulting JSON includes task_type, subtasks[] with agent and depends_on[],
output_format, complexity, recommended_tier, and recommended_model per subtask
(from model_router.get_recommendation).
Tier Dispatch
recommended_tier is director.py's TASK_TIER_MAP value — a legacy 1/2/3
complexity shorthand that picks the dispatch
mechanism below, NOT the provider tier. AGENT_ROUTING[<agent>] in
openrouter_wrapper.py still names per-agent NIM/Groq/Ollama tier bindings
(e.g. python-specialist/research-analyst/data-specialist at NIM/Tier B+,
writing-specialist at Groq/Tier B) — those bindings are dormant, not
deleted, under Anthropic-only (directiva usuario 2026-08-18): see
.claude/rules_db/archive/multi-tier-dormant-2026-08.md. Today, route class 1
and class 3 work to Sonnet directly (finance-specialist/code-reviewer's
Sonnet/Opus bindings are the only ones still live). Canonical tier table:
.claude/rules/03_tiering_and_routing.md.
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.
- 10d ago First seen · 97 lines · 3 tokens per session scan A e9aabff42e7b
orchestrator is an agent published in the GitHub repository senda-labs/DQIII8 (11 stars, last pushed 21d ago), licensed MIT. It adds 3 tokens to every session and 1,037 once invoked, about $0.0000 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
evolution-manager
Analyzes evaluation history and proposes harness modifications for next-session evolution.
router
Classifies tasks via 6-axis taxonomy and selects optimal harness from pool.
synthesizer
Merges independent codebases from ensemble worktrees into a single working project via file-level comparison, cherry-pick, and verified integration.
adversarial-review
Self-attacking implementation executor. Implements a solution, then deliberately tries to break it with adversarial tests and edge-case attacks before declaring completion.
deep-interview
Clarification-first executor. Resolves ambiguous requirements through structured interviews, builds a confirmed spec, then executes against it.
divide-and-conquer
Decomposition executor. Splits large tasks into independent sub-tasks, solves each in isolation, then integrates and verifies the combined result.