Repository navigation
fix: keep discovered agents on memory v1 until owner opt-in - #11630
Conversation
|
Intent: Keep automatically discovered ordinary agents on Global Memory V1. Create private V2 only through explicit member creation or the existing owner opt-in action. The maintainer requested immediate submission and no further local tests. Remaining verification runs in GitHub CI; no CI or review requirement is waived. |
Design Review (Fable 5) — 🟡 CONCERNSDesign-level review of Design-Verdict: CONCERNS Unexplained Watch
Suggestions
[DESIGN-REVIEWED] f2f5a54 |
Opus 4.8 Review — ✅ no blocking findingsReviewed Review detailsConfirmed: No findings. [OPUS-REVIEWED] f2f5a54 Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
GPT 5.6 Review — ✅ no blocking findingsGPT 5.6 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
First Principles Review (Fable 5) — ✅ human override accepted@bolichen97 overrode this lane for |
|
span=5cfc52ae0386
|
|
span=4bac962a3a96
|
|
span=ac4a39fc9293
|
|
span=971703c52762
|
|
/ai-review override first-principles 7168d7b: Maintainer decision, explicitly authorized by bolichen: ordinary discovered and reinstalled agents must default to shared Global Memory V1, not automatically receive private V2. Private V2 requires explicit member creation or owner opt-in. The shared-memory trade is intentional; existing V2 bindings and isolation remain unchanged. This approves the product-default decision only and does not waive CI or other review gates. |
Human judgment recorded@bolichen97 marked the first-principles AI finding as false positive, not applicable, or explicitly accepted for
This decision applies only to this commit. A new push requires a new judgment. |
UX Review (Fable 5) — ⏭️ skippedRevision |
f7db15b to
1793bea
Compare
|
/ai-review override first-principles 1793bea: Maintainer-authorized decision: discovered agents default to shared Global V1; V2 requires explicit creation or opt-in. V2 is unreleased development work, so old V2 data compatibility is out of scope. No stored data is deleted and existing isolation checks remain. This carries the same approved decision to this SHA; other CI and review gates still apply. |
|
AI-review override not recorded: keep the reason to 500 characters or fewer. |
Human judgment recorded@bolichen97 marked the first-principles AI finding as false positive, not applicable, or explicitly accepted for
This decision applies only to this commit. A new push requires a new judgment. |
|
span=6f3b1c923d19
|
|
span=7a5aa49dc697
|
1793bea to
f5dfd42
Compare
|
/ai-review override first-principles f5dfd42: Maintainer-authorized decision: discovered agents default to shared Global V1; V2 requires explicit creation or opt-in. V2 is unreleased development work; old V2 data compatibility is out of scope. This revision only syncs the existing fix onto the upstream test-harness repair. Stored data and isolation checks remain unchanged; all other CI and review gates still apply. |
Human judgment recorded@bolichen97 marked the first-principles AI finding as false positive, not applicable, or explicitly accepted for
This decision applies only to this commit. A new push requires a new judgment. |
f5dfd42 to
f2f5a54
Compare
|
/ai-review override first-principles f2f5a54: Maintainer-authorized decision: discovered agents default to shared Global V1; V2 requires explicit creation or opt-in. V2 is unreleased; old development-data compatibility is out of scope. This revision only resolves a test-file conflict against main, preserving both regression assertions and upstream worker tests. Stored data and isolation checks remain unchanged; all other review and CI gates apply. |
Human judgment recorded@bolichen97 marked the first-principles AI finding as false positive, not applicable, or explicitly accepted for
This decision applies only to this commit. A new push requires a new judgment. |
Backport #11565, #11456, #11630 and #11594 onto the existing release hotfix branch. Track runtime descendants, recover witnessed process groups, preserve V1 discovery defaults, and reap abandoned Linux scopes. Use the first-parent delta of e66570d; retain release session APIs, existing memory bindings and conservative cleanup ownership checks. Carry source regressions and add release-specific binding and hook tests. No workflow, dependency, version or live configuration changes.
Backport #11565, #11456, #11630 and #11594 onto the existing release hotfix branch. Track runtime descendants, recover witnessed process groups, preserve V1 discovery defaults, and reap abandoned Linux scopes. Use the first-parent delta of e66570d; retain release session APIs, existing memory bindings and conservative cleanup ownership checks. Carry source regressions and add release-specific binding and hook tests. No workflow, dependency, version or live configuration changes.
Problem / Motivation
Ordinary agent discovery creates private Memory V2 without an owner choosing it. Opening chat can therefore opt an installed agent into private memory, or fail when private execution is unavailable.
Why it matters
Discovering an installed agent should not change its memory model. Private memory must be an explicit choice, not a side effect of opening chat.
What changed (motivation → approach → change)
🟩 added · 🟨 changed · 🟥 removed · 🟦 unchanged
Discovery in
dashboard/handlers/agents.pynow only registers the agent. It no longer allocates a private store, publishes store ownership, or cleans up a newly allocated store. The existing explicit creation and opt-in paths retain all private-memory checks. Existing bindings and package-prune archival remain unchanged.Tests cover discovery under unsupported private execution and the provisioning pause, repeat sync, explicit opt-in, retained archives, failed publication and concurrent owner edits. Chat selection tests explicitly create their private-member precondition instead of relying on discovery. The ledger deletion-order test uses a clock after the recorded last write, so filesystem and wall-clock skew cannot mask its unlink assertion. Production retention and workflow behavior are unchanged. The memory, crew, dashboard and security module docs describe the changed default.
The requested default is shared Global V1, not memory disabled. A newly discovered ordinary agent can use that shared memory; it does not get a new isolated empty store. Owners who need private isolation must explicitly create or opt into V2. Existing private-memory checks are unchanged, and this change does not migrate or delete stored data.
Maintainer scope: Memory V2 is still in development, not generally released. Compatibility, migration and recovery for old V2 development data are not requirements of this fix. In particular, reusing a removed private member's identity with retained old V2 records is not promised as a supported V1 migration. No retirement metadata API or UI compatibility layer is introduced. The maintainer authorized the shared-V1 discovery default in the approved First Principles product decision, with explicit permission to carry that same decision to subsequent revisions.
API compatibility:
POST /api/agents/syncno longer returns409 member_memory_unavailablefor private provisioning or execution restrictions. It registers ordinary V1 agents instead. Scan and persistence failures still return errors; explicit private creation and opt-in retain their existing refusals. No new UI or configuration flag is introduced.Tests
test_dev_fleet_app.py,test_member_memory_mcp_routing.pyandtest_overload_home_fence.py; their cause is not yet established. They are not reported as a clean gate.Manual verification
Real HTTP test clients exercise discovery followed by the existing explicit private-memory update. No live gateway or user memory was changed. Browser verification has not been performed.
Screenshots / video
No frontend component, layout or string changes. This PR changes the backend discovery default; no browser screenshots were captured.
Related Issues
no linked issue: maintainer-requested correction to the automatic discovery default.
Pattern harvest
Not generalizable: discovery and explicit member creation intentionally have different memory-allocation semantics. Behavioral regression tests pin this boundary without introducing a new lint rule.
Checklist