GitHub CopilotBusabase × GitHub Copilot

Knowledge base for GitHub Copilot

GitHub already records code changes well. Add a reviewed handoff layer for decisions and operational facts that must move between IDE Agent, cloud agent, pull requests, and the systems beyond GitHub.

Release fact proposalIn review
Busabase Inbox reviewing a release fact proposed during a GitHub Copilot workflow
The pull request can merge before a deployment fact is accepted by the system that owns it.

One product, several execution surfaces

The same task can start locally, continue in the cloud, and end in a review queue.

GitHub distinguishes IDE agent mode from Copilot cloud agent. The cloud agent works in an ephemeral GitHub Actions-powered environment and can create a branch and pull request; IDE Agent edits the developer's local environment. Copilot code review is another consumer of repository context. A durable handoff must say which surface produced each claim.

IDE AGENT

Local workspace

Edits and tests on a developer machine.

CLOUD AGENT

GitHub-hosted task

Research, plan, branch, commits, and a pull request.

CODE REVIEW

Pull request

Checks proposed code against repository context and policy.

Repository context already has owners

Do not duplicate GitHub instructions, commits, and pull requests in a second database.

GitHub supports repository-wide instructions, path-specific instructions, AGENTS.md, and Copilot Memory. Repository-level memories are scoped to the same repository and validated against cited code before use. These are strong mechanisms for code context. Busabase is for the facts whose owner is not the repository alone.

ArtifactJobCanonical owner
.github/copilot-instructions.mdHow Copilot should build, test, and validate this repository.Git
AGENTS.md / path instructionsInstructions that change with directory or file scope.Git
Commit and pull requestThe exact code change and its review discussion.GitHub
Customer, release, support, or operational decisionA fact reused by people and Agents outside one repository.Busabase

The handoff chain

A merged pull request is one event, not the entire delivery record.

A task may begin as an Issue, become an Agent session, produce a pull request, pass CI, deploy, and then change a release or customer-facing state. Each transition has different evidence and a different person responsible for accepting it.

ISSUEIntentWhat outcome was requested?
SESSIONExecutionWhat did Copilot investigate and change?
PULL REQUESTCode reviewWhat code was approved?
DEPLOYMENTRuntime evidenceWhat actually reached an environment?
CANONICAL RECORDOperational truthWhat may downstream teams now rely on?

Release evidence contract

Give every cross-system fact enough structure to survive the pull request.

The record should point back to GitHub rather than copy it. A small schema makes it clear when code is merged but rollout, documentation, support readiness, or customer impact still needs review.

FieldMeaning
work_itemIssue or task that authorized the work
pull_requestReviewed code change
commit_shaExact version evaluated
environmentStaging, production, region, or tenant
verificationTests, health checks, or observed result
operational_stateProposed, ready, released, rolled back, or blocked
ownerPerson accountable for the state
effective_atWhen downstream users may rely on it

Review outside the pull request

Code approval and business acceptance can happen at different times.

Copilot can propose a release-state change after reading the merged pull request and deployment evidence. Busabase shows the exact record fields that would change. Product, support, or operations reviews the consequence before the new state is merged.

01

Code merged

GitHub owns the commit and PR history.

02

Evidence attached

Runtime checks and rollout notes are linked.

03

State proposed

Copilot submits only the fields that should change.

04

Owner accepts

The responsible team merges the operational record.

Events can keep the ledger current

Use automation to propose follow-up work, not to erase the reviewer.

A deployment webhook or scheduled Agent can detect a new event and open a Change Request. Delivery logs show whether the event arrived; the record review still decides whether its interpretation becomes canonical.

deploy.succeededPropose release state and verification URL
deploy.failedPropose blocked state and incident owner
rollback.completedPropose restored version and affected scope
Busabase webhook delivery log supporting a GitHub Copilot release handoff
Busabase webhook delivery log supporting a GitHub Copilot release handoff

MCP differs by execution surface

A connection that works in VS Code is not automatically valid for Copilot cloud agent.

The Busabase connection page covers Copilot Agent mode in VS Code, where a Streamable HTTP server can complete OAuth. GitHub's cloud-agent documentation currently says its MCP integration supports tools only, not resources or prompts, and does not support remote MCP servers that use OAuth. Treat these as separate deployment choices.

VS Code Agent modeStreamable HTTP + OAuthUse the existing connection guide and confirm auth_verify.
Copilot cloud agentRepository MCP configurationSelect only required tools; current remote OAuth limitation applies.
Copilot code reviewShares repository MCP settingsReview tool access because calls may run autonomously.

Questions about Copilot handoffs

GitHub Copilot knowledge base FAQ

Does Busabase replace GitHub Issues or pull requests?

No. GitHub should remain canonical for code tasks, commits, branches, pull requests, and code review. Busabase owns reviewed facts that cross repositories or feed product, support, release, and operations workflows.

Does it replace Copilot Memory?

No. Copilot Memory stores repository facts and personal preferences for supported Copilot features. Busabase is useful when a record needs explicit fields, an external owner, a review decision, and reuse outside the same repository.

Can cloud agent connect the same way as VS Code Agent mode?

Not necessarily. GitHub documents different MCP capabilities and authentication limits for cloud agent. Configure and verify each execution surface independently.

Should deployment events write directly to the record?

A deterministic event may trigger a proposal, but a successful delivery does not always establish business readiness. Keep review when the state affects customers, support, compliance, or coordinated rollout.

What should the first Base track?

Start with one handoff where teams currently disagree: release readiness, migration status, security remediation, incident follow-up, or customer-impacting change state.

Start after the pull request

Turn one merged change into a reviewed fact every team can use.

Connect Copilot in the environment you actually use, define one release-state record, and review the first handoff from GitHub evidence to operational truth.