Engineering Board

Repository pattern memory that helps an engineering agent find a root cause.

The board is the database.

Run /board-demo to make this synthetic example.

This visual shows real deterministic fixture output for the pattern pipeline.

This visual shows real deterministic fixture output for root-cause memory.

This visual shows the tested context-to-outcome memory loop.

This repository uses its own Engineering Board.

Product description

Engineering Board is a repository-owned pattern-intelligence system for engineering agents.

The system records bugs, features, questions, and observations as Markdown evidence. It connects recurring findings in a deterministic graph.

The graph helps an agent investigate a shared cause across different domains. This method reduces repeated corrections of individual symptoms.

Markdown is the canonical record. A pull request can show each change to this record.

BOARD.md, GRAPH.yml, JSON analysis, and HTML are derived views. The system can build these views again from the canonical record.

A hypothesis is separate from a deterministic graph fact. Only investigation evidence or fix evidence can confirm a hypothesis.

The optional tdd → review → validate loop can test a fix. This loop supports the pattern memory, but it does not define the product.

Milestone C adds:

• A transparent investigation score from 0 through 100
• A durable H### hypothesis lifecycle
• Negative memory for a rejected claim.

Each score component is visible. A score does not prove that a cause is true.

Milestone D puts this memory in the agent's decision path:

• board_context retrieves relevant clusters, hypotheses, negative memory, and Learnings before the agent selects a fix.
• Each result shows a stable title, a typed summary of at most 2,000 characters, the epistemic state, structural signals, score components, match reason, and canonical sources.
• board_outcomes records an explicit fix result against an H### hypothesis.
• Structured outcomes update Learning state and confidence through a separate preview and apply operation.
• A derived value report counts verified reuse and systemic fix evidence. It does not count prompts, sessions, or other activity.

Milestone D.1 adds a repository-only evaluation harness and eight sanitized cases. The default contract prepares 24 isolated baseline/context pairs for a Codex reference run. Other clients can run as optional replications without changing the product gate. Protocol and package tests establish supported client surfaces without requiring provider accounts. The first Codex run scored 100 percent in both positive baseline and context arms, so the corpus was retained as a non-scored calibration set. A separate locked evidence corpus excludes declared scoring oracles and requires rejected memory in its lexical-decoy contexts. Its reference run scored 100 percent for context and 83.33 percent for baseline. The 16.67-point difference did not meet the required 25-point improvement. The project does not claim that the context improves agent diagnoses.

The unlocked version 4 proposal now limits each positive case to one visible current incident. It also requires a positive classification to connect that incident to prior repository evidence. A non-scored proposal preflight found that v1.11.0 ranks the expected memory but does not include the memory title, cause, or summary in the returned result. Context contract version 2 now returns that bounded canonical content with its separate epistemic state, match reason, and sources. The proposal remains unlocked. Run a new non-scored preflight before the product owner reviews an exact version 4 baseline. See evaluation/README.md for the proof boundary and operator commands.

Product differences

Some Git boards show visible state but have little analysis. Some memory systems have useful analysis but keep the source outside the repository.

Engineering Board combines these properties:

• Repository evidence that a pull request can review
• A cross-domain pattern graph with evidence links
• H### records with alternatives and falsifiers
• Negative memory for rejected claims
• Atomic claims for parallel agents
• Passive capture
• A Claude Code plugin
• An MCP server.

Native Claude Code Tasks and Engineering Board have different purposes.

Native Tasks store personal task state in ~/.claude/tasks/. This state is not part of a project pull request.

Engineering Board stores shared project memory in the repository. Use Native Tasks for temporary personal work. Use Engineering Board for durable project knowledge.

Install the Claude Code plugin

Add the repository marketplace:

/plugin marketplace add GhostlyGawd/engineering-board

Install the plugin:

/plugin install engineering-board

Set up a board:

/board-setup

/board-setup creates the board structure. It also checks the required permissions.

Run the contained demonstration:

/board-demo

The command creates a synthetic run in .engineering-board/demo/pattern-intelligence/.

The command connects three findings from different domains. It then requests one hypothesis that cites the evidence.

The hypothesis has status: proposed. It includes an alternative explanation and a falsifier.

The report gives an exact cleanup command. The cleanup operation preserves a changed run.

For explicit setup values, use:

/board-init <project> [affects-prefix]
/board-install-permissions

Use the pattern-memory workflow

1. Work in Claude Code.
2. Let SessionStart and UserPromptSubmit retrieve relevant systemic memory.
3. Run /board-context <project> when you want the same bounded brief explicitly.
4. Read the memory title, typed summary, epistemic state, match reason, and sources. Inspect the cited canonical record before you rely on a proposed cause.
5. Let the Stop hook capture a finding in engineering-board/<project>/_sessions/.
6. Run /board-promote to preview canonical changes.
7. Apply the unchanged promotion plan after you review it.
8. Run /board-insights <project> when you need the complete ranked investigation view.
9. Run /board-hypothesis <project> propose to preview an H### record.
10. Review the evidence, alternative, counter-evidence, confidence basis, and falsifier.
11. Apply the unchanged hypothesis plan.
12. After verification, run /board-outcome <project> preview ....
13. Apply the unchanged outcome plan. Review and apply each returned Learning plan separately.

An outcome records held, failed, partial, or inconclusive. It can confirm, weaken, reject, or leave a hypothesis unchanged only through a compatible explicit disposition.

Use /pm-start only for advanced batch promotion. Use the optional Worker loop only when you want to test a selected fix.

Use one session mode

/pm-start and /worker-start set .engineering-board/session-mode.json.

A session can have one mode. Start a new session to change the mode.

On Claude Code web, each session uses a new clone. On a local installation, the mode file stays on disk.

To return to passive capture:

1. Start a new session.
2. Read the SessionStart message.
3. If a mode remains, delete .engineering-board/session-mode.json.

Register the MCP server

Register the PyPI package with the Claude Code command-line interface (CLI):

claude mcp add engineering-board -- uvx engineering-board-mcp

To run the server from a clone:

git clone https://github.com/GhostlyGawd/engineering-board
claude mcp add engineering-board -- python3 "$(pwd)/engineering-board/mcp-server/engineering_board_mcp.py"

For Claude Desktop, add this object to claude_desktop_config.json:

{
  "mcpServers": {
    "engineering-board": {
      "command": "uvx",
      "args": ["engineering-board-mcp"]
    }
  }
}

mcp-server/README.md also contains setup procedures for Codex CLI, Gemini CLI, and Cursor.

The plugin registers the same server through .mcp.json.

Product surfaces

The plugin has four session modes:

The canonical Stop procedure is hooks/stop-hook-procedure.md.

Commands (21): /board-setup, /board-demo, /board-context, /board-outcome, /board-promote, /board-pattern, /board-insights, /board-hypothesis, /board-run, /board-init, /board-rebuild, /board-graph, /board-view, /board-remember, /board-pause, /board-resume, /pm-start, /worker-start, /board-install-permissions, /board-claim-release, and /board-migrate.

Agents (8): board-manager, finding-extractor, consolidator, tidier, learnings-curator, tdd-builder, code-reviewer, and validator.

Skills (5): board-intake, board-triage, board-resolve, board-consolidate, and board-insights.

Hooks (4 events): SessionStart, PostToolUse(Write), UserPromptSubmit, and Stop.

The MCP server has 19 tools. All tools use the same canonical Markdown and the same deterministic core.

Architecture boundary

Canonical cards, hypotheses, Learnings, and BOARD-ROUTER.md use Markdown. Derived views include BOARD.md, GRAPH.yml, context briefs, value reports, JSON, and HTML.

The product does not require SQLite. A future SQLite index must be disposable and rebuildable. Measured query requirements must justify it.

The runtime uses Claude Code, bash, and python3. It has no runtime package dependency.

Read ARCHITECTURE.md for the full system map.

Roadmap boundary

Milestone D ships in v1.11.0.

The cross-session Conductor remains a draft RFC. /board-run <entry-id> ships only the single-session inner loop.

Cross-repository intelligence, hosted services, and a required database remain outside the current product boundary.

docs/PRODUCT_EVOLUTION_SPEC.md is the authoritative product-direction source.

Contribute

Run the complete test suite:

bash tests/run-all.sh

The run-all command uses the maintained suite list. Read CONTRIBUTING.md before you change the repository.

Support

• Use GitHub Discussions for questions.
• Use GitHub Issues for bugs and features.
• Use GitHub Security Advisories for a vulnerability.
• Use the live board for current project work.
• Use GitHub Sponsors to support the project.

GhostlyGawd maintains this open-source project.

License

The project uses the MIT License.

Language status

The owner approved the current controlled-English text. The project does not claim formal ASD-STE100 compliance, certification, or independent review.