Ralph Orchestrator is a framework that repeatedly runs AI-agent tasks until they finish or reach an iteration limit. Developers use it to coordinate autonomous coding work through command-line, web-dashboard, and MCP-server interfaces, with state managed per workspace. The catalogue entries provide agent skills, agents, instructions, and plugins for operating Ralph.
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 skills add mikeyobrien/ralph-orchestrator --skill robot-interactiongit clone --depth 1 https://github.com/mikeyobrien/ralph-orchestratorWrote 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/skills/mikeyobrien/ralph-orchestrator/robot-interaction)<a href="https://agentmods.dev/skills/mikeyobrien/ralph-orchestrator/robot-interaction"><img src="https://agentmods.dev/badge/skills/mikeyobrien/ralph-orchestrator/robot-interaction/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/skills/mikeyobrien/ralph-orchestrator/robot-interaction"><img src="https://agentmods.dev/badge/skills/mikeyobrien/ralph-orchestrator/robot-interaction.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 | $0.00013 | $0.00451 |
| Opus 5 | $0.00006 | $0.00226 |
| Sonnet 5 | $0.00003 | $0.00090 |
| Haiku 4.5 | $0.00001 | $0.00045 |
Grade A, and why
robot-interaction 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.
What it actually says
Human Interaction (RObot)
A human is available via Telegram. You can ask blocking questions or send non-blocking progress updates.
Hard rule: Do NOT send progress and human.interact in the same turn.
If you need to ask, include the status in the question and skip the progress update.
Progress updates (non-blocking)
Send one-way notifications — the loop does NOT block:
ralph tools interact progress "All tests passing — starting integration phase"
Use for: phase transitions, milestone completions, status updates, FYI messages.
Asking questions (blocking)
Emit human.interact with your question — the loop blocks until the human replies or timeout:
ralph emit "human.interact" "Decision needed: [1 sentence]. Options: (A) ... (B) ... Default if no response: [what you'll do]"
Always include:
- The specific decision (1 sentence)
- 2-3 concrete options with trade-offs
- What you'll do if no response (timeout fallback)
The human may also send proactive guidance at any time (appears as ## ROBOT GUIDANCE in your prompt).
When to ask (blocking)
- Ambiguous requirements that can't be resolved from context
- Irreversible or high-risk decisions (deleting data, public-facing changes)
- Conflicting signals where you need a tiebreaker
- Scope questions (should I also do X?)
When NOT to ask
- Routine implementation decisions you can make yourself
- Status updates — use
ralph tools interact progressinstead - Anything you can figure out from specs, code, or existing context
- Rephrasing a question already asked this session
Rules
- One question at a time — batch related concerns into a single message
- After receiving a response, act on it — don't re-ask
- If guidance contradicts your plan, follow the guidance
- Prefer
progressfor FYI messages; reservehuman.interactfor decisions that need input
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 · 55 lines · 13 tokens per session scan A fd6f7595001f
robot-interaction is a skill published in the GitHub repository mikeyobrien/ralph-orchestrator (3,133 stars, last pushed 2d ago), licensed MIT. It adds 13 tokens to every session and 451 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 2026-09-10.
Other skills, from other repositories
chart
Use when the user asks to visualize data with charts, graphs, or plots using the chart tool (bar, line, scatter, pie, time series, etc.).
writing-tests
Write unit tests, component tests, and integration tests for AiderDesk using Vitest and React Testing Library. Use when creating new tests, adding test coverage, configuring mocks, setting up test files, or debugging failing tests.
playwright-testing
Playwright E2E testing patterns, cross-browser configuration, page objects, and CI setup. Use when creating E2E specs, visual regression suites, or configuring Playwright in CI. Trigger terms: playwright, e2e, trace, page object, cross-browser.
agent-memory
Creates, queries agent expertise profiles in AGENT-EXPERTISE.md; increments file-familiarity counters after each task; ranks candidate agents by recency, task-area match. Use when deciding which agent should handle a file, checking who last worked on a module, recording task outcomes, or assigning work based on past…
data-engineering
Transforms, validates, loads data in ETL pipelines. Use when building scrapers, validating NDJSON feeds, or importing data into CMS/DB targets.
ralph-linear
Fetch Linear tickets and build them with ralph-starter.