BBriefOpen app

Agent-native engineering system

Give every coding agent the same map, loop, and review bar.

Brief does not replace your agent or harness. It sits beside Codex, Claude, Cursor, terminal agents, local harnesses, and MCP IDEs, then makes each one inherit your repo map, rules, skills, design systems, reference implementations, and proof bar.

Connect an agentSee the loop
TerminalVS CodeClaude CodeCodex CLICursorLocal harnessesAny MCP client
brief-agent-runlive contract
Codex CLIuse Brief on APP-218 and fix safely
Claude Codefrom now on, use Brief before serious edits
Cursor@brief inspect existing primitives first
Terminalreview this diff before handoff
Local harnessuse the team's Brief loop before editing
terminalbrief.start -> map -> mission -> review -> handoff
contextReceipt.packetId: ctx_9f1amapCoverage: application-sourceruntimePacket.status: readyreview.verdict: readyproofBundle.status: ready

How Brief gets useful

Connect the agent. Teach the team loop. Prove the work.

Brief is not a prompt library or another destination app. It is the team-owned operating layer that makes existing coding agents and custom harnesses use the same repo map, wiki, rules, examples, review bar, and proof trail.

01
Connect

Install once in the agent

Codex, Claude, Cursor, VS Code, terminal, IDE agents, and local harnesses keep their normal UI. Brief verifies the active repo before serious work starts.

brief.install_status
02
Teach

Turn repo knowledge into defaults

Brief learns repo rules, skills, commands, examples, design systems, prior receipts, and AutoWiki docs so the agent starts with your team's loop.

brief.initialize_repo_wiki
03
Run

Start work in normal words

Paste a ticket, PR, incident, screenshot, or task. Brief routes context, map, wiki, mission contract, review gates, handoff proof, and reusable learning.

brief.start
What it learns
Learn installed workflowsbrief.skill_learning

Reads repo guidance, skills, commands, hooks, and Cursor rules, then extracts triggers, gates, stop conditions, proof artifacts, and reusable run contracts. A workflow like `/tpr TPR-660` becomes a Brief-native ask such as `Use Brief on TPR-660`.

Make Brief defaultbrief.auto_engage

Returns the default instruction, smoke test, learned-workflow bridge, and install snippets for AGENTS.md, CLAUDE.md, Cursor rules, or the repo-local `.brief/wiki` artifact.

Provide referencesbrief.save_reference

Save a known-good helper, component, route, PR, workflow, design pattern, or harness recipe with source paths, use-when guidance, and do-not-copy guardrails. Agents call `brief.reference_implementations` before creating another primitive.

Why agents get better
Reuse insight

Flags new helpers, components, routes, and keys that overlap mapped shared primitives before they become parallel code.

Test insight

Suggests co-located and policy-relevant tests for the exact files and risk surfaces the agent touched.

Setup insight

Warns when the connector is aimed at the wrong repo, when context was captured too late, or when cloud sync is not carrying team memory.

What the connector gives the agent

The useful objects appear inside the coding session.

brief.start

Start from the user's words

A ticket, PR, bug, screenshot, incident, or plain repo task becomes the right loop without asking the user to memorize tools.

01inferred run contract
brief.codebase_map

Map the repo before the agent edits

Routes, services, API handlers, data models, tests, owners, observability, and reusable primitives.

02source coverage
brief.initialize_repo_wiki

Install a repo wiki agents will read

A connected agent writes `.brief/wiki`, source-cited pages, AGENTS/CLAUDE hooks, freshness manifests, and continuous refresh workflow templates.

03repo-local docs
brief.mission_plan

Run loops instead of prompt chains

Feature work, ticket fixes, runtime incidents, hard bugs, stale PRs, flaky CI, internal apps, docs, experiments, and reviews get bounded.

04parseable contract
brief.review_diff

Review against the team's bar

Bug bar, security pass, required tests, mapped risk, redundant primitive warnings, secrets, team rules, comment budget, and cross-agent challenge.

05review receipt
brief.save_personalization

Make good taste reusable

Personal skills, team styles, design systems, approved examples, repo scopes, guardrails, and promotion receipts across surfaces.

06team-share proposal

Mission lab

Compose the work like a console command.

Pick the source, runtime signal, loop, and review bar. Brief turns it into the same run contract a coding agent can execute and a teammate can audit.

generatedbrief.run_loop --source selected --runtime selected --recipe selected --review-bar selected --proof required
source
runtime
loop
review

you choose a source, runtime signal, loop, and review bar

brief attaches map, wiki, team style, runtime facts, stop conditions, validation, and review gates

agent executes the smallest safe diff, then saves handoff proof

Run contract

The chosen controls become a mission plan with context receipts, validation requirements, review gates, and a handoff.

mapevidencemissionreviewhandoff

High-leverage work becomes first party

Brief is strongest where agents usually lose the plot.

stale_pr / merge_conflict / ticket_fix

Finish stuck work

A stale PR, half-done branch, merge conflict, or blocked ticket still has useful intent.

GitHubLinearJiraCIgit
incident_fix / bug_investigation / flaky_test

Fix from runtime facts

A Sentry issue, Datadog trace, log query, CI failure, alert, or incident points to a real failure.

SentryDatadoglogstracesCI
qa_edge_cases / flaky_test / pr_review

Ratchet quality upward

A flaky test, weak coverage path, recurring review comment, or edge-case gap keeps costing the team.

test historyreview memorycode mapAutoWiki
internal_app / experiment / performance

Ship internal tools safely

The team wants an analytics, admin, or operations app without inventing auth, DB, or access rules from scratch.

repo conventionsDB schemaauth rolesaudit logs
session_library / session_review / save_personalization

Turn great runs into team memory

A strong agent session, loop recipe, or coding style should become reusable across repos.

receiptshandoffsvalidation logsteam review
AutoWiki viewersource-grounded pages
OverviewArchitectureTech StackProject StructureEntry PointsValidationTeam StandardsReview Memory

Generated pages are not brochure docs. They are the map your agents can cite: architecture, request paths, validation targets, standards, owners, loop usage receipts, and the `.brief/wiki` files installed into the repo.

Review gatesignal before human review

Bug bar

Prioritize defects that are introduced by the diff, actionable, and worth fixing.

Depth control

Use shallow checks for fast WIP feedback and deeper review for shared or security-sensitive code.

Security pass

Trace secrets, injection, auth, data exposure, and supply-chain risk before handoff.

Comment budget

Keep findings signal-dense by surfacing the highest-impact issues first.

Cross-agent challenge

Record which agent authored the diff and which independent surface reviewed it.

Keep your coding surface

One connector pattern. Native habits stay intact.

The setup goal is boring: connect Brief where tools are supported, use the loop recipe where they are not, then keep writing normal tasks in the agent or harness your team already likes.

MCP direct

Terminal

Repo shell

Call brief.install_status, then brief.mission_plan.
MCP direct

VS Code

VS Code MCP server settings

Use Brief before editing this workspace.
MCP direct

Claude Code

Claude Code MCP servers

Write the loop for this task with brief.mission_plan.
Loop recipe

Claude console

Project instructions or chat preface

Follow this research-plan-validate-review loop for the task.
MCP direct

Codex CLI

Codex MCP connector settings

Call brief.context_for_task before changing files.
MCP direct

Cursor

Cursor MCP config

Use Brief's loop and review gates for this change.
MCP adapter or recipe

Local harnesses

Project package, adapter, or startup prompt

Use Brief as the team context layer for this task.
MCP or recipe

Cursor console

Composer/chat instructions

Follow the loop, cite sources, and run review before handoff.
MCP direct

Any MCP IDE

IDE MCP server list

Start with install_status and mission_plan.
Direct connector

Paste one MCP config and let the agent run the team loop in the repo.

Claude CodeCodex CLICursorLocal harnessesAny MCP client
Engineering stack

Route the tools high-velocity product teams already trust into the same governed agent loop.

GitHubLinearSentryDatadogSlackCI
Delegated agents

Keep the same context, loop, and receipt standard when work moves to cloud coding agents or review bots.

DevinGitHub CopilotCodeRabbitSourcegraph / Amp

Local MCP first, cloud sync for derived metadata when approved.

Works in Terminal, VS Code, Claude Code, Codex CLI, Cursor, local harnesses, and any MCP-capable IDE.

Linear, Jira, GitHub Issues, Sentry, Datadog, CI, logs, traces, and internal data contracts can become mission evidence.

Every final update can carry source paths, validation, review verdict, artifact proof, remaining risk, and receipt ids.

Ready path

Connect Brief once. Keep working in your agent.

Connect agent