Motir

• 2026年10月15日Planning bug: a re-plan's lay reconciles existing children but schedules no author session for them, so a child whose title changed (or whose body the requirement falsifies) keeps its stale body, and the plan still says "reconciled"
• 2026年10月15日Prepare the word-list review sheet — the words read from the open pull request at a named commit, their counts against the 100 × 100 floor, each adjective's work-item key, the app's shipped languages, and a flagged-pair shortlist screened from every pairOnboarding that matches motir.co — the Motir look and "vibe the project" voice, a GitHub connect that does not lose you, no waiting on the index, a generated first-project name, and import proven against real Jira, Linear and Plane
• 2026年10月15日Guide me through reads the project's code before every turn — even to draft or save a list — runs out of turns, discards what it read, and hands the person code reading and command-running steps instead of confirming them itself
• 2026年10月15日Planning bug (both planners): an author session that finds the parent story's criterion false never amends the story with update_item — sibling cards each pick a side and ship opposite behaviours
• 2026年10月15日(motir-core · design) The DISCREPANCY row in the run's findings strip — a `run-modal--discrepancy-finding.mock.html` delta of `design/runs` § What the run PRODUCED: its marker and hue, its label and point count, where it sits beside bugs and plans, its door to the card comment, and its place in the node's count
• 2026年10月15日(motir-core · decision) A run builds through a wrong HOW and records a DISCREPANCY — the decision page amending `run-findings-protocol.md` and `dispatch-run-record.md`: the one test between build-through and STOP, the `discrepancy_recorded` finding the server writes from `record_run_discrepancy`, its pointer-only data, one record per leg, and what stays STOP
• 2026年10月15日Planning bug: the smoke-scale run was laid as a CHILD of the story whose completion closes it, against that story's own placement hand-off
• 2026年10月15日The job tables stop growing without bound — a retention pass for succeeded job_queue rows (and job_step where it is unbounded) that never deletes a row a live reader, the DLQ or the code-graph lookups need, beside MOTIR-6935's job_run passmotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test
• 2026年10月15日Integration test of the retention pass against the real database: what survives, the watermark, dedup inside the window, batch bound and backlog drainmotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test
• 2026年10月15日Record the new dedup horizon in the job-queue decision recordmotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test
• 2026年10月15日Retention service and the daily system.job-queue-retention job reporting rows deleted per tablemotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test
• 2026年10月15日Retention repository methods: batched delete of old succeeded job_queue rows, their job_step rows and orphaned job_event rows, with window and batch constantsmotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test
• 2026年10月15日Audit every job_queue reader and record each one's keep-rulemotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test
• 2026年10月15日A person reading a plan or an approval never sees the page refresh under them — every router.refresh() and whole-page re-render on the plan, approval and Workbench surfaces replaced by an in-place update of only what changedmotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test
• 2026年10月15日(motir-core) The To approve and Planning tabs apply a nudge in place — their rows read from the one endpoint's approvals and planning windows into client state, merged by identity, no route refreshmotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test
• 2026年10月15日(motir-core) Story E2E with acceptance video — a reader through a 50-write MCP burst on a generating plan, a planned plan, an approval and every Workbench tab: zero _rsc refetches, zero document loads, place preserved, Approve keeps the decided planmotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test
• 2026年10月15日(motir-core) Story integration test gate — coverage floor, appliers driven from a fake signal source, rows/counts agreement, render-count and DOM-mutation assertionsmotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test
• 2026年10月15日(motir-core) An architecture guard fails when router.refresh, useCoordinatedRefresh or window.location is reachable from the plan, approval and Workbench live-update pathsmotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test
• 2026年10月15日(motir-core) The open approval reads the seam, never refreshes its page — subject-moved notice unchanged, port scroll, selection and half-typed refusal text survive every update, over /workbench and over an item pagemotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test
• 2026年10月15日(motir-core) Approve lands without a route refresh — the planning hosts reconcile the decided plan from the approve response and the review read, and the Workbench counts behind the overlay are current when it closesmotir-core's web tier scales horizontally under agent traffic — each added `app` instance adds throughput (~500 req/s per instance, toward ~10k active agents), connections stay bounded, no reader is refreshed, measured first and proven by a re-runnable load test

更早

透過 Atom 訂閱