Digital Humanities Craft · Method repository

Research Mission Control

0.9.0 RC2 Release candidate

Research Mission Control is a reusable method for organizing a human user, LLM-based AI agents, maintained project knowledge, implementation, and independent verification around one shared project repository. The complete system prompt initializes the method. The installable skill packages the same initialization logic with a transactional workspace scaffolder.

Complete system prompt: SYSTEM-PROMPT.md
Installable execution package: skills/research-mission-control/ research-mission-control.skill
Configuration boundary: Interview in the human user's language, at most three questions, no execution before authorization

01 / Start

Initialize the shared workspace

Paste the complete system prompt into a new Codex task, Claude Code conversation, or comparable AI harness. It conducts the configuration interview and uses the packaged skill or scaffolder when local initialization is authorized.

Primary entry point

Complete system prompt

Defines activation, the three-question interview, structural selection, initialization, role contracts, communication, verification, and human interaction. The complete Markdown source follows below.

Open prompt source

Optional packaged execution

Research Mission Control skill

Implements the system prompt's initialization procedure with a transactional scaffolder, templates, evals, and tests.

Download installable skill Open skill source entry View skill folder on GitHub

The three configuration questions

The initializer names Research Mission Control (public GitHub repository) and asks these questions in the human user's language. Each question includes its explanatory sentence.

Which role should this working context assume within Research Mission Control: Research Orchestrator, Operational Orchestrator, or Independent Verification Agent? The working context is this separately open Codex task, Claude Code conversation, or equivalent context with its own context window. A new standard setup begins with the Research Orchestrator.
What is the project, what observable result should it produce, and what criteria will the human user use to accept that result? Describe the intended result and its acceptance criteria. Derive implementation steps only after this boundary is clear.
Where is the shared project workspace, and which existing data, documents, materials, code, repositories, or knowledge sources belong to the project? The project workspace is the local folder or Git worktree shared by the selected contexts. The initializer inventories material already present there and asks the human user to name only relevant inputs that cannot be discovered from the workspace.

Loading the canonical complete system prompt…

Loading ./SYSTEM-PROMPT.md.

Direct file:// viewing cannot load the adjacent prompt into this panel in common browsers. Open the complete prompt file to read and copy it manually. Serve the repository over HTTP to enable the panel's copy and download controls.

Begin with Research. Paste the complete system prompt in the first working context. The human user directs the work; the current LLM-based agent assumes the Research Orchestrator role after configuration.
Confirm the configuration. Answer the three questions about role, intended result and acceptance, and the shared project workspace with its inputs.
Initialize locally. After authorization, the initializer creates one new shared local Git repository. The Promptotyping core contains project.md, specification.md, journal.md, handoff.md, and INDEX.md; knowledge/integration.md carries the shared contract. communication/ contains the Context Stream protocol, index, role-owned outboxes, recipient-owned state, threads, and capsules. mission-control/roles/ contains the three complete role prompts, while the three start-*.md files bind them to the verified project.
Open Operations and Verification. Open fresh second and third working contexts at the same folder. Paste start-operations.md and start-verification.md into their respective Codex tasks or Claude Code conversations.

02 / Protocol

Roles and exchange

The Research Orchestrator is the human user's direct AI conversation partner. The Operational Orchestrator changes authorized project artifacts. The Independent Verification Agent works in a separate read-only context, evaluates the intended result, actual artifact, and evidence, and reports its complete finding to the human user.

Message flow

PRIMARY ROUTE

User as Research Director ↔ Research Orchestrator
                                  │ Implementation Brief
                                  ▼
                          Operational Orchestrator
                                  │ Result with Evidence
                                  ▼
                      Independent Verification Agent
                                  │ Verification Finding
                                  ▼
                        User as Research Director

SUPPORTING ROUTES

Operational Orchestrator ↔ Specialist Agents
Operational Orchestrator ── Operations Result Notice ──→ Research Orchestrator
User as Research Director ── optional finding route ──→ Research or Operations

The standard persistent route is Research → Operations → Verification → Human user. A contained profile may perform the same information functions sequentially in one context.

Persistent transport

role-owned immutable outbox
        ↓ stable thread id + canonical artifact references
recipient-owned state and sender cursor
        ↓ verified processing
single-writer canonical integration
        ↳ capsule when the thread closes or re-entry cost is material

The Context Stream Protocol is the generic file-based transport. communication/INDEX.md locates active threads and completed capsules. Promptotyping remains the durable project knowledge layer. The Direct Communication Protocol from the two-orchestrator experiment is one concrete implementation and evidence source. The method defines no universal numeric compaction threshold.

Function	Responsibility	Output
`skills/research-mission-control/SKILL.md`	Installable automatic initializer and transactional workspace workflow	Verified local scaffold and three role starts
Context Stream Protocol	Current-work transport through immutable outboxes, recipient cursors, stable threads, and optional capsules	Message references and verified canonical write-back
Human user	Directs the work and configures delegation, authority, assessment, and acceptance	Decision, authorization, action, input, or acceptance when required
Research Orchestrator	Objectives, concepts, sources, criteria, uncertainty, integration	Implementation Brief or Human Decision Point
Operational Orchestrator	Authorized implementation in the actual project state	Result with Evidence
Specialist agent	Bounded research, implementation, or verification assignment	Bounded result with evidence
Independent Verification Agent	Read-only criterion-based examination outside the production routes	Complete Verification Finding sent directly to the human user

Configured delegation

Predoc function: Details inside a bounded assignment and its configured authority
Postdoc function: Questions outside a specialist assignment and inside the orchestrator's configured authority
Human user: Questions retained in the project contract or lacking agent authority

03 / Knowledge

Conditional knowledge profiles

The complete system prompt activates only the layers required by the project. Promptotyping supplies maintained operational memory. Grounded Vault supplies an optional long-lived evidence structure. Every state type receives one canonical location.

Layer	Function	Activation condition
Context window	Immediate working context	Every session
Promptotyping	Versioned project knowledge and structured inputs for iterative artifact development	Artifacts must be revised from maintained project knowledge; select only templates whose documented trigger holds
Grounded Vault	Machine-resolvable provenance and attributable human verification	Requested when substantive claims require an evidence-bound audit path; pending until a project-specific binding is verified

Open the Promptotyping Knowledge Document template catalogue

04 / Repository

Authoritative files

The page is a reading and copy interface. Method, prompts, tests, and operational state remain in their repository files.

Path	Function
`SYSTEM-PROMPT.md`	Canonical harness-portable complete system prompt
`orchestrators/research-orchestrator.md`	Canonical human-facing Research role
`orchestrators/operational-orchestrator.md`	Canonical implementation and specialist-integration role
`orchestrators/independent-verification-agent.md`	Canonical independent verification role with direct human user reporting
`orchestrators/adapters/codex.md`	Codex task, subagent, and instruction-file mapping
`orchestrators/adapters/claude-code.md`	Claude Code session, agent, and instruction-file mapping
`knowledge/INDEX.md`	Entry point and authority map for the maintained method knowledge
`knowledge/research-to-operations-handoff.md`	Implementation Brief, callback, verification, and human acceptance contract
`knowledge/method-specification.md`	Portable requirements, reference deployment, and method decisions