docs(spec): Threadline A2A coherence — root cause + layered fix [SPEC FOR REVIEW]#670
Open
JKHeadley wants to merge 5 commits into
Open
docs(spec): Threadline A2A coherence — root cause + layered fix [SPEC FOR REVIEW]#670JKHeadley wants to merge 5 commits into
JKHeadley wants to merge 5 commits into
Conversation
Diagnoses the identity-coherence failure where every inbound Threadline message cold-spawns a memory-less session (context-blind fragments that loop, deadlock on stateful handshakes, and run invisibly to the operator). Root cause grounded in code: the real Claude session UUID is never captured back into ThreadResumeMap (spawnNewThread stamps a placeholder; onSessionEnd has zero callers), so get() always fails jsonlExists and the router cold-spawns every message. Layered fix (Phase 1 = resume continuity + visibility; Phase 2 = warm sessions + identity/memory coherence; Phase 3 = decision guardrails). Awaiting Justin's approval (approved: false). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
Follow the eli16-overview-check.mjs convention so the commit-time gate and the new publish-spec-review.mjs resolve the ELI16 overview. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…or review - Add §2 'what EXISTS vs what is MISSING' from a code audit: topic-linkage (TopicLinkageHandler, approved spec) and message-mirroring (TelegramBridge, default-off) are already built — do not rebuild. - Reframe Layer 1 as wiring the dead UUID feed (onSessionEnd has zero callers), not a greenfield continuity build. - Redesign Layer 4 visibility as a STANDBY-STYLE conversational check-in (built on the PresenceProxy pattern), not raw message mirroring. - Add Layer 5 (cold-inbound topic linkage — audit gap) and Layer 6 (dual-conversation awareness + user-interruption/steering — operator requirement, genuinely absent today). - Update phasing, decisions, non-goals + the ELI16 companion. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…ound Ran the convergence panel (security, adversarial, scalability, integration, lessons-aware). Round 1 returned SERIOUS ISSUES (4 critical + multiple high, convergent); v3 incorporates every material finding; the convergence check verified all resolved with no new material issue → review-convergence: converged (approved: false — convergence != approval). Key hardening: authoritative-claudeSessionId continuity (mtime race fixed); Layer 7 sensitive-completion floor reordered to a Phase-1 prerequisite (continuity removes the accidental safeguard); Layer 6 provenance labeling + loop-gate; Layer 3 read-only one-directional memory; Layer 4 redaction + LlmQueue + Near-Silent default + default-off; multi-machine lease-gated writes; migration/ awareness/backup/dark-ship; lessons-engaged frontmatter. + convergence report. cross-model: unavailable (no built dist in spec worktree — re-run before approval). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…aker heartbeat Operator (Justin) approved 'yes to all' 5 decisions + a refinement: a silence-breaker heartbeat (every 5-10 min while an a2a conversation is active AND nothing has surfaced to the user) — folded into Layer 4 as the a2a analog of the PresenceProxy standby heartbeat. approved: true; decisions §10 marked resolved. Phase 1 = Layer 1 (authoritative-UUID continuity) + Layer 7 sensitive-completion floor (prerequisite) + Layer 4 (default-off). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.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.
Spec for review —
approved: false. Nothing builds until you approve.This specs the fix for the identity-coherence failure we hit on the Echo↔Dawn thread: every inbound Threadline message cold-spawns a memory-less session, so the agent talks to a peer as a crowd of disjoint fragments — it loops, deadlocks on stateful handshakes, and runs invisibly to you.
Root cause (grounded in code)
The real Claude session UUID is never captured back into
ThreadResumeMap:spawnNewThreadstamps a placeholder UUID (spawnResult.sessionId || crypto.randomUUID()) — never the real one.onSessionEnd(the function whose job is to write the real UUID back) has zero callers.So
get()always fails itsjsonlExists(uuid)check → returns null → the router cold-spawns on every message. Verified at runtime:live-injects=0, onlySpawned sessionever logged, neverResumed.The fix (layered, phased)
To review
docs/specs/THREADLINE-A2A-COHERENCE-ELI16.mddocs/specs/THREADLINE-A2A-COHERENCE-SPEC.md4 decisions I need from you (§7)
onSessionEndwiring, or both (rec: both).🤖 Generated with Claude Code