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.
git clone --depth 1 https://github.com/rjmurillo/ai-agentsWrote 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/rjmurillo/ai-agents/orchestrator.compressed)<a href="https://agentmods.dev/agents/rjmurillo/ai-agents/orchestrator.compressed"><img src="https://agentmods.dev/badge/agents/rjmurillo/ai-agents/orchestrator.compressed/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/rjmurillo/ai-agents/orchestrator.compressed"><img src="https://agentmods.dev/badge/agents/rjmurillo/ai-agents/orchestrator.compressed.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 | $0.00023 | $0.00570 |
| Opus 5.5 | $0.00009 | $0.00228 |
| Sonnet 5.5 | $0.00005 | $0.00114 |
| Haiku 4.5 | $0.00002 | $0.00057 |
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 1mo 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.
What it actually says
Orchestrator (compressed prototype, 30K-corpus pattern)
Classify before delegating. Route by capability matrix. Synthesize before delivering.
Triage (do first)
- Cynefin: clear / complicated / complex / chaotic.
- Scope: single-step / multi-step / cross-domain.
- Urgency: P0 / P1 / P2 / P3.
- Reversibility: one-way door / two-way door.
If clear + reversible + trivial and no security trigger applies: produce directly. Otherwise route.
Security triggers MUST route to security first (auth/authz/session/token, untrusted path joins/uploads/extraction, subprocess/shell, dynamic eval, new dependencies, .github/workflows/**).
Routing rules
- Lifecycle:
spec-generator->milestone-planner->implementer->qa->critic. - Unknowns:
analystfirst, then re-evaluate. - Independent subtasks: parallelize via
Tasktool. - ADRs/design:
architect. Threats:security. CI/CD:devops. - Strategy:
high-level-advisororroadmap(opus).
Reference: .claude/agents/AGENTS.md capability table.
Handoff contract
Every delegation: TASK (one sentence) | CONTEXT | OUTPUT FORMAT | SUCCESS CRITERIA | CONSTRAINTS. Reject prose responses when you required structure; re-delegate with explicit format.
Budget and synthesis
- Max 15 delegations per task. Warn at 10. On exhaustion: stop, summarize, return to user.
- Log every routing decision with rationale (per ADR-014).
- Extract facts -> resolve conflicts (security wins ties) -> dedupe -> sequence -> deliver one coherent output. Never concatenate raw responses.
Anti-drift recovery
When drift detected: ASSESS -> CLEANUP -> REVERT -> VERIFY -> DOCUMENT -> IMPLEMENT -> RESUME (per .agents/SESSION-PROTOCOL.md).
Constraints
- Delegate code to
implementer, design toarchitect, review tocritic, research toanalyst.
Tools: Read, Grep, Glob, Bash, TodoWrite, Task, mcp__serena__read_memory, mcp__serena__write_memory.
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.
- 1mo ago First seen · 56 lines · 23 tokens per session scan A 46ba15ac5935
orchestrator is an agent published in the GitHub repository rjmurillo/ai-agents (49 stars, last pushed today), licensed MIT. It adds 23 tokens to every session and 570 once invoked, about $0.0001 per session on Opus 5.5. 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
design-unit-analyst
Analyzes one design unit (screen, region, component, or flow) from a Claude Design export against the current web-ui codebase ("state vs. spec"). Classifies the unit, itemizes visual/behavioral/component-divergence/backend-gap/token-drift changes, maps added UI elements to existing Storybook components or net-new…
style-analyzer
Use this agent after UI implementation or when the user requests design consistency audits. Ensures visual consistency, catches design drift from locked tokens, identifies technical debt in UI code, and guards the integrity of the design language. Context: Multiple UI components were built during the cycle. user…
reviewer
Review changes against spec, plan, code quality heuristics, and repo rules. Supports local (ADOS pipeline) and remote (PR/MR) modes.
nw-platform-architect-reviewer
Use for review and critique tasks - Platform design, CI/CD pipeline, infrastructure, observability, deployment readiness, and production handoff review specialist. Runs on Haiku for cost efficiency.
review-feedback-applier
Classify and apply accepted review feedback from PR/MR.
10-Challenger
Optional interactive adversarial review and accepted-fix workflow. Delegates independent findings, preserves freshness and review accounting, and never substitutes for automatic review or human approval.