feat(assistant): human review and owner-approved Ornn publication for agent learning, dormant behind assistant:agent-learning (0.61.0) - #1765
Merged
Conversation
added 8 commits
October 4, 2026 19:33
… dormant behind assistant:agent-learning (L1 PR-1)
# Conflicts: # backend/src/services/feature_flag_service.rs
… request access snapshot Thread creation paid a personal flag resolution (which reads org memberships) on every new thread even when the agent has no learning config, breaking the org group single-auth-resolution guarantee. Check the learning config by _id first and reuse the caller's org access snapshot for the ACL.
…s before any learning read Keeps both guarantees: flag-off enrollment touches no learning collection, and thread creation with a request snapshot adds no org membership reads. The full personal resolution runs only after a learning config exists when the flag may be on broadly.
… agent learning proposals (L1 PR-2) Merges origin/main (0.60.0). Encrypted proposal review, edit and reject with fingerprint suppression; one-use owner action card bound to one publication operation; deterministic NyxID-built private Ornn packages through a fixed operation allowlist; ambiguous publication reconciles and never repeats a started write; publish-and-pin shares the B2 skills revision fence. Org agents refuse publication (owner_binding_unavailable). Ordinary B2 skill saves are unchanged and never touch learning storage.
📊 Code coverage
Gate: line coverage must stay at or above the threshold. Ratchet plan (W21): Backend → 55%, CLI → 50%, Frontend → 30% by quarter end. |
…eview # Conflicts: # Cargo.lock # backend/Cargo.toml # cli/Cargo.toml # cli/src/wizard/bundle-meta/index.hash # frontend/package-lock.json # frontend/package.json
chronoai-kai
added a commit
that referenced
this pull request
Oct 5, 2026
…nostics on failure (#1769) Coverage (Backend) and Coverage (Backend Base) lost mongod at the same cumulative point (~840-960 s, auth_device_service tests) on #1765/#1766 with ~7 GB host memory still free; mongod held ~8 GB when it vanished. Raise the descriptor limit far above the nextest peak (thousands of per-test databases' WiredTiger files stay open in one llvm-cov process) and cap the WiredTiger cache at 2 GB. On failure or cancellation, print the container state (exit code, OOMKilled), mongod descriptor count/limits, kernel OOM lines and the server log tail so any recurrence is diagnosable. Co-authored-by: chrono-kw <chrono-kw@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
L1 PR-2: human review and owner-approved publication for agent learning proposals, dormant behind
assistant:agent-learning(default off). Builds on PR-1 (#1756, v0.58.0).skills_revision, the config revision and the package SHA-256. The card is consumed in the same retrying transaction that claims the publication lease. A replay can only resume its own durable operation.owner_binding_unavailable.Backward compatibility
agent_skill_service::set) are unchanged and never read or write learning storage; a test asserts this. The Ornn allowlist covers every existing B2 read and search path./assistant/nyxagent.finish_runnow uses the retryingand_run2transaction pattern, with unchanged semantics.Validation
RUST_MIN_STACK=1572864: backend filters learning 3, publication 2, review 11, agent skills 11. Review tests cover ACL refusals (org member, guest, non-orchestrator, unrelated agent), card invalidation after edit orskills_revisionchange, reconcile refusal for zero and two matches, no reissue, and lease-expiry resume.--all-targets -D warnings; fmt. Frontend: 4,228 tests, lint, build.setregression (a retry-less transaction plus a dormant learning write), retry-less transactions, UI conventions, the durable error boundary, and 403 on timeouts.