sync(v2.12): infrastructure/stability parity for policy snapshots, embeddings, event-loop init, self-update, and history compatibility #128

New issue

Open

opened 2026-09-09 14:58:54 +00:00 by nsaspy · 0 comments

nsaspy commented

2026-09-09 14:58:54 +00:00

Owner

Upstream

Agent Zero v2.12: https://github.com/agent0ai/agent-zero/releases/tag/v2.12

Missing parity

Ingest the remaining v2.12 runtime/stability changes that are not owned by the focused transport/Responses/provider/security tickets.

Required work

Reuse one fresh scoped-tool-policy snapshot across a resolution batch.
Unify embedding calls behind one batch entry point and remove duplicate transport paths.
Serialize shared event-loop initialization so concurrent first use cannot strand tasks.
Preserve rollback-stash identity during self-update by tracking Git object IDs.
Preserve non-LLM mode through result-metadata dictionary/history round trips.
Preserve legacy assistant-history call compatibility while the v2.12 explicit llm_result contract is adopted in #127.
Share editor log presentation across local and remote tools.
Keep rate-limiter demo/example code out of test discovery.
Use compact 4-space-indented workdir tree extras where relevant to agent prompt output.
Route nudge as a system message and keep message templates minified.
Preserve the simplified one-action WhatsApp QR pairing flow.
Update stale Launcher README links to the reviewed Launcher release line.
Ignore .venv in generated gitignore templates.

Acceptance

Concurrent first-use runtime tests cannot deadlock/strand the shared loop.
Embedding callers traverse one tested transport/batch abstraction.
Self-update rollback restores the exact stashed Git object state.
Non-LLM result metadata survives serialization/history round trips.
Legacy supported history call forms remain covered during the #127 migration.
Test discovery excludes demos; generated gitignore contains .venv; QR pairing remains one-action.
Focused regression/performance tests demonstrate no policy-resolution or embedding-call regressions.

## Upstream Agent Zero v2.12: https://github.com/agent0ai/agent-zero/releases/tag/v2.12 ## Missing parity Ingest the remaining v2.12 runtime/stability changes that are not owned by the focused transport/Responses/provider/security tickets. ## Required work - Reuse one fresh scoped-tool-policy snapshot across a resolution batch. - Unify embedding calls behind one batch entry point and remove duplicate transport paths. - Serialize shared event-loop initialization so concurrent first use cannot strand tasks. - Preserve rollback-stash identity during self-update by tracking Git object IDs. - Preserve non-LLM mode through result-metadata dictionary/history round trips. - Preserve legacy assistant-history call compatibility while the v2.12 explicit `llm_result` contract is adopted in #127. - Share editor log presentation across local and remote tools. - Keep rate-limiter demo/example code out of test discovery. - Use compact 4-space-indented workdir tree extras where relevant to agent prompt output. - Route nudge as a system message and keep message templates minified. - Preserve the simplified one-action WhatsApp QR pairing flow. - Update stale Launcher README links to the reviewed Launcher release line. - Ignore `.venv` in generated gitignore templates. ## Acceptance - Concurrent first-use runtime tests cannot deadlock/strand the shared loop. - Embedding callers traverse one tested transport/batch abstraction. - Self-update rollback restores the exact stashed Git object state. - Non-LLM result metadata survives serialization/history round trips. - Legacy supported history call forms remain covered during the #127 migration. - Test discovery excludes demos; generated gitignore contains `.venv`; QR pairing remains one-action. - Focused regression/performance tests demonstrate no policy-resolution or embedding-call regressions.