Why: Hiding a tool declaration alone does not prevent a model from calling a tool already in the runtime registry.
Evidence: The agent-native loop continues after tool errors; the plan now requires a local stop latch before any later provider call.
Impact: Clarify the zero-paid test and implementation boundary without changing the approved design or authorizing model usage.
Refs: Agent-System-Documents#95; Agent-System-Dev#512
Why: The approved written design needs testable work items before changing model behavior or spending another sample.
Evidence: The plan maps all three design duties to existing code seams, RED-to-GREEN checks, a fixed combination gate and a separately authorized real-model exit.
Impact: Record stage-five review material without modifying product code, using paid models, or opening merge gates.
Refs: Agent-System-Documents#95; Agent-System-Dev#512
Why: The approved model samples showed unsupported approval claims, repeated source input and no final answer before the run budget stopped.
Evidence: The design cites the preserved A01 provider usage, persisted findings and independent sample audit without claiming new model validation.
Impact: Define a reviewable source, claim and final-reserve revision while keeping code, paid trials and merge gated.
Refs: Agent-System-Documents#95; Agent-System-Dev#512
Why: The controlled trial made one more paid provider request after rejecting a nonliteral quote.
Evidence: Independent transcript audit found a 14231-token continuation within the same child run; original receipts and failed quality verdict remain intact.
Impact: Make the authorization deviation explicit and reference a zero-paid prevention guard; no additional model run is authorized.
Refs: Agent-System-Documents#95; Agent-System-Backend#193; Agent-System-Dev#512
Why: The approved v7 sample saved two checkpoints but still missed semantic and final-result acceptance.
Evidence: Independent DB replay, provider usage reconciliation, source checks and the original failed evaluation receipt.
Impact: Preserve the exhausted one-shot boundary and keep all draft PRs unmerged.
Refs: Agent-System-Documents#95; Agent-System-Dev#512
Why: The approved extra sample ended with no valid saved findings and exhausted its per-run token threshold.
Evidence: Provider responses, strict-source rejections, Usage and independent analysis are retained; v7 has only local checks.
Impact: Define a separately approvable two-run 30000-token-per-run experiment without altering original acceptance rules or claiming model success.
Refs: Approved U2/U3 plan r0.1; independent extra A01 assessment
Why: The project and proposal indexes still said no U2/U3 implementation or experiment existed.
Evidence: Current combination, full PG backend, two HTTP browser cases and independent code review are linked with their actual statuses.
Impact: Documentation only; real-model acceptance is explicitly failed and no release is claimed.
Refs: Approved U2/U3 task card and validation report
Why: Keep the latest merged candidate reviewable without turning failed model findings into a release claim.
Evidence: Exact d06792c composition, two current browser cases and independent model review are linked and source-labeled.
Impact: Documentation only; model quality gate and merge remain closed.
Refs: Approved U2/U3 task card
Why: The model produced no admissible findings, so release and further paid work remain gated.
Evidence: Original provider calls, 42203 tokens, independent SQLite replay and strict source rejection are linked from the validation report.
Impact: Correct planning status and recommend a bounded lower-complexity test design for separate approval.
Refs: Approved U2/U3 plan r0.1; extra A01 user approval 2026-10-13
Why: Preserve other approved planning records and index updates.
Evidence: Exact current main fetched; no overlapping source behavior.
Impact: Retain both planning histories.
Refs: Agent-System-Dev U2/U3 delivery
Why: Make the current corrected candidate reviewable while preserving the failed model acceptance boundary.
Evidence: Native SELF receipt for exact source composition 7fd5288 is green; independent reviewer verifies Backend 6a2d85c and all fixed regressions.
Impact: Documentation only; the six failed or limited model samples remain unchanged and no merge or deployment is claimed.
Refs: Approved U2/U3 plan; pending explicit extra A01 test authorization
Why: Planning-only status no longer matches the candidate, while failed model samples must remain explicit.
Evidence: Native SELF receipts and independent REPLAY reports distinguish code checks, six model cases and the failed quality/fidelity boundaries; relative links verified.
Impact: Update only this approved U2/U3 planning record and validation report; CI, merge, deployment and an extra sample remain pending.
Refs: User confirmation 2026-10-13; Dev feature-fast-slow-u2-u3-plan
Why: The 2026-10-13 login recovery flow was approved and merged
(Backend #187, Frontend #76, Dev #504); the master proposal must carry
the three adjudications, implementation record, and evidence pointers
per the repo's lineage conventions.
Evidence: Dev #504 HEAD CI (test/test-pg/svc-realcombo) success; browser
acceptance receipts be7ae9a2/b3a9e691 bound to Dev HEAD 268ef29 (SELF).
Impact: r3 is an implementation record only; R/KTD/U numbering semantics
unchanged; U7 explicitly NOT marked complete (atomic bootstrap and CLI
credential delivery remain). INDEX.md status line updated.
Refs: Dev docs/svc-deploy/2026-10-13-login-recovery-contract.md
Why: the 2026-10-12 drill left artifact recovery failed, and the retest changed that SELF result without changing the write-traffic gate.
Evidence: backup-20261013-013354 on Dev a83588dd and backend 46073d30; K1-K7 true, RPO 65s, RTO 3s, five artifacts matched and cross-user access returned 404. SELF.
Impact: proposal moves to r2.3. Second-instance writes stay closed until REPLAY or CI and the second host exist.
Refs: docs/records/2026-10-12-u7-ops-drill-report.md docs/proposals/2026-10-11-collaboration-storage-and-distribution.md
Co-authored-by: Cursor <cursoragent@cursor.com>
Why: make persistent activity and grounded findings reviewable before business changes.
Evidence: fixed source audit and scoped plan self-review; business and live checks remain NOT_RUN.
Impact: proposed contracts, implementation sequence, compatibility and bounded verification only.
Refs: fast-slow design r1.1, Dev feature/fast-slow-u2-u3-plan
Why: Keep planning progress aligned with actual implemented corrections without duplicating CI and merge state.
Evidence: Reference UI155, Frontend210 and 19 local browser cases via Dev native receipts; final GATE remains separately tracked.
Impact: Planning metadata only; no deployment or model-routing claim.
Refs: work-companion-ui r0.2/r1, UI01-UI13
Why: Keep the approved design and implementation plan consistent with the current candidate.
Evidence: Reference actual SELF verification and pending independent Review and GATE, without copying command output.
Impact: Planning progress only; inherited live-model capabilities and deployment remain separate.
Refs: UI01-UI13, work-companion-ui U5
Why: preserve concurrent planning records before the Dev combination sync.
Evidence: fetched authoritative Documents main f76cce9e32.
Impact: planning record integration without rewriting history.
Refs: work-companion-ui U5.
Why: B6, C4, and C5 now have same-host results, and the plan still said the ops drills were not done and that replay was the next step.
Evidence: Dev results pgbench-20261012-131948, fault-20261012-142734, and backup-20261012-153441 (SELF). The user declined a capacity replay on 2026-10-12.
Impact: Second-instance writes stay closed. K4 remains a product defect. Replay and C3 are not started.
Refs: feature/u7-ops-drills
Co-authored-by: Cursor <cursoragent@cursor.com>
Why: implement a simple workbench for nontechnical users under continuous authorization.
Evidence: current source audit and the reviewed r0.1 prototype; runtime verification remains pending.
Impact: design and plan only; task execution state stays in the Dev task card.
Refs: UI01-UI13 and the existing fast-slow display design r1.1.
Why: the 2026-10-12 batch-update failure still had identity, rollback, and real-combo gaps after Dev #474, Backend #180, and RDM #276. The recovery plan is the single record for that work, and the rdmcand data-ownership check needed a written result before any restore.
Evidence: read-only queries on runtime-backups/{dev,rdm1,rdmcand}/20261012T133921Z and the matching domain backups. No business rows were moved.
Impact: adds r3.1 sections D1/D2/D3 and the ownership finding that dev conversations predate rdmcand's use of dev/data, so no cross-instance restore is indicated. r3's earlier unapproved text stays unapproved.
Refs: Dev feature/svc-d1-identity-flow; Backend feature/d1-identity-open
Co-authored-by: Cursor <cursoragent@cursor.com>
Why: keep r2/r3 proposal updates rebase-friendly before PR.
Evidence: origin/main at fetch time.
Impact: Documents feature tip includes latest main.
Refs: feature/svc-account-durability-plan
Why: Capacity ladder/60m/2h dual-instance windows finished; planning
progress must reflect SELF-pass vs remaining GATE/REPLAY/C3 gaps.
Evidence: SELF load results summarized in the new capacity report.
Impact: Preferred next becomes independent REPLAY or C3/C4/C5.
Refs: Dev feature/u7-capacity-wrap-measure; release-multi-instance.md
Co-authored-by: Cursor <cursoragent@cursor.com>
Why: Dev #451 cannot pin modules/documents while feature tip and main diverge.
Evidence: merge-tree reports submodule conflict 1048e7c vs 9e4ee27.
Impact: Documents #84 tip includes main@1048e7c plus r1.1/U1 plan.
Refs: Dev PR #451; Documents PR #84
Why: distinguish timed observations from current service ownership.
Evidence: SELF task-sync and read-only observations at 18:14 Beijing time.
Impact: planning metadata only; runtime and component deployment unchanged by this task.
Refs: account r2; recovery r3
Why: account help, dev credential writes and crash recovery need a current, reviewable contract.
Evidence: SELF source and sanitized runtime observations; seven document review lenses; business and fault validation NOT_RUN.
Impact: revise two planning documents only; preserve prior approvals and mark additions pending review.
Refs: Dev feature/svc-account-durability-plan; account r2; recovery r3
Why: User authorized stage-5 start; record plan status for executors.
Evidence: Chat authorization "开始"; no runtime tests.
Impact: Documents proposal status only.
Refs: 2026-10-12-fast-slow-channel-display-u1-plan.md
Why: Record approved stage-4 design with R03 splits and U1 exit
whitelist, plus stage-5 U1 implementation plan for review.
Evidence: Document edits only; no runtime verification.
Impact: Documents proposals INDEX; no business code.
Refs: feature/fast-slow-display-design
Co-authored-by: Cursor <cursoragent@cursor.com>
Why: Delivery requires writing planning progress back after U7
landed; the proposal still claimed U1+U2 start / NOT_RUN.
Evidence: SELF doc edit against archived U7 cards and release
checklist A1-A7/B1 facts.
Impact: Readers see code-done vs capacity-window-pending; adds
HTTP multi-instance vs Manager-test lesson.
Refs: Dev feature/u7-capacity-wrap-measure; release-multi-instance.md
Co-authored-by: Cursor <cursoragent@cursor.com>
Why: Capture nil-card panic and intent data-flatten field fixes found
during local prod acceptance without revising the final closure verdict.
Evidence: Backend#173 #174; prod action checklist
Impact: Field-fix log lists the two hotfixes and their PRs.
Refs: GenUI prod deploy; fix track
Why: Record that the known-boundary production deploy/acceptance ran
on the local prod svc instance with auth and PG.
Evidence: scripts/svc.sh prod list; acceptance API checklist
Impact: Usage guide boundary list no longer says deploy is pending.
Refs: GenUI prod deploy task
Why: user approved the proposal and U1/U4 specs and authorized push/PRs
plus U1/U4 implementation start.
Evidence: SELF conversation approval 2026-10-12; implementation still NOT_RUN.
Impact: planning status only.
Refs: docs/proposals/2026-10-12-svc-unified-deploy-and-login-key.md
Co-authored-by: Cursor <cursoragent@cursor.com>
Why: capture the reviewed design for binary/source install, env prep,
and passwordless activation via one-time restricted sessions.
Evidence: SELF source review against Dev af9c2de; implementation NOT_RUN.
Impact: Documents planning index only; no runtime change.
Refs: docs/proposals/2026-10-12-svc-unified-deploy-and-login-key.md
Co-authored-by: Cursor <cursoragent@cursor.com>
Why: policy requires updating applicable plan progress on task
completion; WP3-I3 landed the admin-zone entrance via the generic
declaration mechanism.
Evidence: Dev PR #414 and live verification comment (R5, SELF).
Impact: progress line of 2026-10-11-rdm-two-trees-trace-audit-integration.md only.
Refs: WP3-I3/KTD4.
Why: Record the independent U5-style review for the first two B-class
surface actions and update the usage guide allowlist note.
Evidence: proposal r1 status approved; usage doc API table updated.
Impact: Documents authority for backend whitelist implementation.
Refs: U5 action-scope r2; plan GenUI boundaries 2026-10-12
Co-authored-by: Cursor <cursoragent@cursor.com>
Why: The direction closes at U7 - the master record needs its
companion usage guide (enable/generate/revise/act/disable for owners)
and the combined evaluation record covering every landed increment
with its evidence chain and known boundaries.
Evidence: Live E2E on dev@b634a27 (real MiniMax-M3): progress page
presented autonomously, conversation surfaces=2, project aggregation
returns cross-conversation pages under the owner scope; receipts in
Dev task feature/generative-ui-u7-acceptance.
Impact: Documents-only (new guide + index closeout).
Refs: r2 master record; U5 scope r2; Dev task
feature/generative-ui-u7-acceptance
Why: between the review round and the approval decision another session completed and merged svc-autorecover (PR #396/#400/#401/#402), so the (a)-decision tracking must reflect the landed slice and the remaining U2 delta.
Evidence: SELF - merged main history c622372 and archived card .task/archive/feature-svc-autorecover.md (10/10 tests, 146/146 governance, CI green on e0b49d1).
Impact: review record, INDEX; proposal r2 body unchanged.
Refs: PR #396#401
Why: the human gate resolved on 2026-10-12 - r2 approved (five stages at once) and option (a) adopted; the review record, proposal index and status navigation must reflect it.
Evidence: SELF - user decision relayed in conversation; stage-review details in the Dev task card; proposal r2 body unchanged (SHA256 9cfda252...).
Impact: review record, INDEX entry; no plan content change.
Refs: Dev task card .task/feature-service-restart-recovery-plan.md
Why: the reviewer record must reflect that all r2 gap claims hold at the post-sync backend pin 25fbea85, and that the svc-autorecover ownership decision now has 373 lines of unregistered, partially failing in-flight work to weigh.
Evidence: SELF - drift-check receipt in the Dev task card (exit 0, all cited gaps hold); svc-autorecover suite observed 5 of 10 tests failing on 2026-10-11.
Impact: review record addendum only; proposal r2 (SHA256 9cfda252...) unchanged.
Refs: Dev task card .task/feature-service-restart-recovery-plan.md
Why: absorb the external source-review round - pin the shutdown budget change, close the U1 acceptance gap for program/cwd changes, gate U4 public states on a mapping table, and make the svc-autorecover coordination an explicit human decision.
Evidence: SELF - external review round 2 in the review record; r2 SHA256 9cfda2520b913e6966c68fb3f66fb7bc8382a8f7a2b69cb364122216dcdb72cd; all reviewed claims re-verified against the pinned baselines.
Impact: documents only; four targeted amendments (U1/U3/U4/section 9) plus version bump and INDEX update; requirements and metrics unchanged.
Refs: Dev task card .task/feature-service-restart-recovery-plan.md
Why: U5 landed via RDM PR #213 (merge 1dce69d); all five implementation units now complete.
Evidence: RDM PR #213 merged=True.
Impact: plan status + U5 progress + INDEX.
Refs: RDM PR #213
weichen
weichen