Fix data race on shared *Agent.historyProvider by not mutating the interface field mid-run

[PratikDhanave]

What

handleHistoryProviderConflict cleared the configured history provider by writing a.historyProvider = nil in the middle of a run, while historyProviderForSession reads that same interface field unsynchronized on every run. The Agent struct has no mutex and historyProvider is a plain interface value, so a shared *Agent run concurrently (each call with its own service-managed session) races the write against the reads — a two-word interface value written and read without synchronization, which go test -race flags and which can produce a torn read.

A shared *Agent is the intended concurrency unit here (sessions are per-conversation and the repo ships a concurrent agent-workflow example), so this is a real data race.

Fix

• Keep historyProvider immutable after construction.
• Add an historyCleared atomic.Bool to Agent.
• handleHistoryProviderConflict now calls a.historyCleared.Store(true) instead of niling the field.
• historyProviderForSession returns nil early when a.historyCleared.Load() is set.

This preserves the existing .NET/Python clear-on-conflict semantics: once a provider promotes a session to service-managed and the conflict policy allows clearing (throw disabled, keep disabled), the configured history provider is cleared globally for subsequent runs. The behavior is identical; only the mechanism changes from mutating a shared interface field to setting an atomic flag, which removes the race.

Testing

• go build ./..., go vet ./agent/..., go test -race ./agent/... all pass.
• Added TestAgent_Run_HistoryProvider_ConcurrentConflictClearIsRaceFree in agent_test.go: it configures an agent with an explicit HistoryProvider, AllowHistoryProviderConflict=true, KeepHistoryProviderOnConflict=false, and a provider whose Run promotes the session to service-managed mid-run, then launches many goroutines calling Run concurrently and drains each stream. Under -race it reports a data race against the pre-fix code and passes after the fix. The existing clear-on-conflict test remains green.

[Copilot]

Pull request overview

This PR removes a data race in Agent when a shared *Agent is run concurrently by preventing mid-run mutation of the historyProvider interface field. Instead, it preserves the existing “clear-on-conflict” behavior by recording the global clear using an atomic.Bool flag that is safe to read/write concurrently.

Changes:

• Added historyCleared atomic.Bool to Agent and stopped mutating Agent.historyProvider during runs.
• Updated historyProviderForSession to disable history provider usage once historyCleared is set.
• Added a concurrent -race regression test covering the clear-on-conflict path.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
agent/agent.go	Replaces racy historyProvider field mutation with an atomic “cleared” flag and gates provider usage on that flag.
agent/agent_test.go	Adds a concurrency test intended to exercise the previous race pattern under go test -race.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[github-actions]

Parity Review: ✅ No cross-repo consistency issues

This PR fixes a data race in the Go *Agent by replacing a mid-run mutation of the historyProvider interface field with an atomic.Bool (historyCleared) flag.

Scope assessment: Internal-only change.

• No exported functions, types, methods, fields, or options were added, removed, or modified.
• The observable behavior — clearing the configured history provider once a session is promoted to service-managed, with AllowHistoryProviderConflict=true and KeepHistoryProviderOnConflict=false — is preserved identically.
• The PR description confirms alignment with .NET/Python clear-on-conflict semantics.

Labels: public-api-change is not warranted (no exported API surface changed). parity-approved is already present and correct — this fix keeps Go semantically aligned with upstream implementations.

No parity issues found.

Warning

Firewall blocked 1 domain

The following domain was blocked by the firewall during workflow execution:

• awmgmcpg

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "awmgmcpg"

See Network Configuration for more information.

Generated by Go API Consistency Review Agent · 18.7 AIC · ⌖ 5.55 AIC · ⊞ 5.9K · ◷