Repository navigation
v2.49.0
What's Changed
- docs(handoff): §101 — the three generations of CodeQL alerts, and why the suppression did not work by @Ikalus1988 in #3449
- fix(automation): three reds, one cause (D1's free tier) and one standing defect by @Ikalus1988 in #3450
- fix(backup): export table by table — --no-schema alone does not dodge the FTS5 refusal by @Ikalus1988 in #3452
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3466
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3467
- docs(handoff): §102 — the round's closing ledger, and why it can be archived by @Ikalus1988 in #3468
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3472
- feat(lessons): adopt eleven bounty lessons, and add "video" to the domain vocabulary by @Ikalus1988 in #3485
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3487
- fix(codeql): keep generated lesson artefacts out of the secret query by @Ikalus1988 in #3488
- docs(codeql): correct which tool produced the sixteen secret alerts by @Ikalus1988 in #3490
- fix: the six verified severe defects from #1775 by @zsxh1990 in #3478
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3492
- fix(heal): refuse to write a fixture whose signature carries a credential by @Ikalus1988 in #3493
- test(site): lock Quickstart copy to data-i18n and keep zh from shipping English by @zsxh1990 in #3475
- feat: govern homepage Quickstart commands and tool count with a site-facts SSOT by @zsxh1990 in #3479
- docs(start): wire Quickstart track A to /start and back-link to Quickstart (fix #1896) by @zsxh1990 in #3473
- feat(quickstart): copy buttons per code block — #3491 resolved against the site-facts SSOT by @Ikalus1988 in #3494
- docs(handoff): §103 — the adoption batch, the sixteen alerts that were not CodeQL, and the Quickstart review by @Ikalus1988 in #3495
- docs(handoff): §104–§106 — what the three §103.6 items actually measured by @Ikalus1988 in #3496
- fix(mcp): the preflight schema says it needs a Bearer token, not just its error by @Ikalus1988 in #3497
- docs(triage): the intake backlog sweep, and why 26 issues sat open after being fixed by @Ikalus1988 in #3498
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3499
- fix(docs,worker): align the intro's claims with the installer, the worker and the fields by @Ikalus1988 in #3500
- lessons: convert intake #3339 (proxy the shell never inherited) and #3470 (a disable nobody verified) by @Ikalus1988 in #3501
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3502
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3503
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3507
- fix: a degraded read must not be reported as an empty corpus by @Ikalus1988 in #3512
- docs(handoff): §111 — what the 00:00 UTC reset verified, and one correction by @Ikalus1988 in #3513
- chore(deps): update pymdown-extensions requirement from >=12.1 to >=12.3 in the pip group by @dependabot[bot] in #3505
- chore(deps-dev): bump wrangler from 4.149.0 to 4.151.0 in the npm group by @dependabot[bot] in #3504
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3522
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3540
- docs(handoff): 2026-10-19 volume — queue state and what the red checks mean by @Ikalus1988 in #3542
- perf(worker): mem the corpus per isolate (opt-in), and decide the D1 question by @Ikalus1988 in #3545
- perf(worker): stop querying D1 for counters for a minute after one failure by @Ikalus1988 in #3547
- fix(worker): a traffic counter is D1 plus KV, not D1 or KV by @Ikalus1988 in #3549
- fix(activity): a day the store refused is drawn as missing, not as zero by @Ikalus1988 in #3551
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3556
- site + mcp: the audit batch (six PRs that turned out to be one stack) by @Ikalus1988 in #3560
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3561
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3562
- docs(handoff): 116 — the audit batch, what the subagents overturned, and what is still a decision by @Ikalus1988 in #3570
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3578
- fix: the stdio receipt must say where the report went, and the preview button was dead by @Ikalus1988 in #3585
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3589
- chore(deps): bump aiohttp from 3.14.3 to 3.14.4 by @dependabot[bot] in #3586
- feat(site): one source for the design tokens (#3541, batch V) by @Ikalus1988 in #3591
- feat(site): the hand-written pages use the vocabulary (#3541, batch W) by @Ikalus1988 in #3592
- fix(worker): the account was spending 8x its free-tier row-read budget on two public endpoints by @Ikalus1988 in #3590
- docs(worker): the CDN serves a cached entry to a request that carries credentials by @Ikalus1988 in #3594
- feat(site): the halos, the colour bars and the purple (#3541, batch X) by @Ikalus1988 in #3593
- feat(site): the icons are Lucide SVG, not emoji (#3541, batch Y) by @Ikalus1988 in #3596
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3597
- docs(handoff): record which PR carried each batch of #3541 by @Ikalus1988 in #3598
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3600
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3606
- feat(site): one accent, one background, one footer (#3541, batch Z) by @Ikalus1988 in #3615
- docs(handoff): the four pitfalls that are already red on main (#3541) by @Ikalus1988 in #3617
- fix(worker): the account was still paying 95% of its rows for one public aggregate by @Ikalus1988 in #3618
- docs(handoff): the D1 row-read round, measured before and after by @Ikalus1988 in #3619
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3620
- docs(handoff): the deployed hour, measured — and where the index lives by @Ikalus1988 in #3621
- docs(handoff): retract the 2.95x claim, and measure today's D1 with traffic as the control by @Ikalus1988 in #3631
- feat(site): the generated pages join the design gate, and the search gate stops reading a stale local index by @Ikalus1988 in #3634
- docs(handoff): §124.10 — the site has two faces, and Cloudflare took 13 minutes by @Ikalus1988 in #3635
- fix(site): the records that were wrong, and a pixel check that can see the generated tree by @Ikalus1988 in #3636
- fix(infra): a build that never started is a failure, not silence by @Ikalus1988 in #3637
- docs: two surfaces publish
freshness, and they are not the same measure by @Ikalus1988 in #3638 - chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3643
- fix(site): the onboarding dialog behaves like a dialog, and says its name in the reader's language by @Ikalus1988 in #3646
- fix(fatal-guard): the crash payload carries the error text, not the exit code by @Ikalus1988 in #3647
- docs(integrations): the connector's own page answers the ownership question, and it is not stale by @Ikalus1988 in #3648
- docs(handoff): §126-§127 — two issue lines, and a missing lock that a flaky label hid by @Ikalus1988 in #3649
- chore(deps): bump certifi from 2026.7.22 to 2026.10.19 by @dependabot[bot] in #3650
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3669
- fix: lock the profile on Windows instead of yielding unlocked by @zsxh1990 in #3538
- fix(pixel-gate): the default tolerance is a measured floor, and a command to re-measure it by @Ikalus1988 in #3670
- fix(profile): lock the backfill write too — the read path was writing unlocked by @Ikalus1988 in #3671
- docs(handoff): §128 — four PRs, and three places where I was wrong about what I knew by @Ikalus1988 in #3674
- fix(fatal-guard): one payload contract on every platform, and a program path with spaces by @Ikalus1988 in #3672
- feat(crash): a crash looks for an existing lesson before it leaves a draft (#3642) by @Ikalus1988 in #3676
- test(fatal-guard): the Windows tokenizer test asserted a token count the machine decides (#3672 follow-up) by @Ikalus1988 in #3678
- docs(handoff): §129 — the crash path, my own mutation audit, and the PR I merged with a red leg by @Ikalus1988 in #3679
- feat(pixel-gate): a criterion for changes that are meant to move the page by @Ikalus1988 in #3680
- fix(pixel-gate): the page was still on its way when the capture called it stable by @Ikalus1988 in #3681
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3683
- fix(pixel-gate): a capture that never settled is not evidence by @Ikalus1988 in #3684
- docs(pixel-gate): the limitation section was measuring a defect that is now fixed + handoff §130 by @Ikalus1988 in #3685
- feat(install): one generic command first, and a client table generated from data by @Ikalus1988 in #3687
- docs(handoff): §131 — the registration form moved, and the three things it taught by @Ikalus1988 in #3695
- feat(site): the registration form moves to /register/, and the homepage says one line by @Ikalus1988 in #3694
- docs(field-report): the external trial audit and the OpenClaw follow-up, and the issues they became (#3689 #3690 #3691 #3692 #3696 #3699 #3700) by @Ikalus1988 in #3703
- fix(worker): the search query is redacted before it reaches storage (#3692) by @Ikalus1988 in #3702
- fix(field-reports): the gate was rejecting a correct report — the live endpoint has served 8 tools since #3259 by @Ikalus1988 in #3705
- fix(setup): the hook redacts the query before it leaves the machine (#3692) by @Ikalus1988 in #3707
- fix(setup): the npm copy of the rule text says redaction is done by code (#3692) by @Ikalus1988 in #3708
- fix(setup): the npm description says what it writes, what leaves this machine, and the one opt-in exception (#3689) by @Ikalus1988 in #3710
- feat(setup): the install says what it is about to write, before it writes it (#3690) by @Ikalus1988 in #3709
- test(hook): the fetch test races the hook's own four-second stall guard (#3692) by @Ikalus1988 in #3718
- docs(handoff): §132–§137 — the trial feedback, the four issues it became, and what the closure turned up by @Ikalus1988 in #3721
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3723
- ci(setup): the setup package is released with the repository, and drift is a red gate (#3691) by @Ikalus1988 in #3715
- docs(handoff): §138 — the tool forgot, which is a different kind of drift by @Ikalus1988 in #3739
- fix(setup): the release moves the version literal, not just the manifest (#3691) by @Ikalus1988 in #3738
- fix(release): bound the commit scan, so the release PR can be rebuilt again by @Ikalus1988 in #3741
- fix(setup): --verify answers two questions, and stops calling a file it cannot write "missing" by @Ikalus1988 in #3743
- fix(release): the extra-files of a package with a path are resolved against that path by @Ikalus1988 in #3744
- chore(data): sync lessons.json + refresh feed by @Ikalus1988 in #3747
- fix(release): restore the commit scan's floor, and gate it by @Ikalus1988 in #3749
- lessons(contrib): answer four claimed questions (#3564 #3299 #3301 #3380) by @zsxh1990 in #3682
- fix(mcp): local stdio search: memory_context ImportError, domain filter, "usage" routing by @vishalhabib99 in #3515
- feat(quickstart): report qs_view/qs_copy/qs_link step events to D1 by @zsxh1990 in #3532
- feat(heartbeat): report helpful-source panel votes over retrievals by @zsxh1990 in #3534
- fix(tests): stop sharing one sandbox across lesson-contract tests by @zsxh1990 in #3530
- feat(intake): post conversion receipts and label ready-to-close (#3509) by @zsxh1990 in #3565
- docs(handoff): §139–§141 — the release could not run, the file was listed but not moved, and the verdict had two questions in it by @Ikalus1988 in #3746
- fix(docs): the quickstart-events page needs the site's own partials by @Ikalus1988 in #3753
- feat(guard): two more mechanical shapes — an internal artifact, a new top-level directory by @Ikalus1988 in #3754
- fix(release): the scan's floor is judged by the release tag, not by the ledger by @Ikalus1988 in #3752
- chore: release main by @Ikalus1988 in #3451
New Contributors
- @vishalhabib99 made their first contribution in #3515
Full Changelog: v2.48.0...v2.49.0
装上它只要一行(含"不会用命令行"的情况)
已经在用 Claude Code 或 Codex:
npx @misaka-net/misakanet-setup装完把这个助手窗口关掉再打开一次,然后随便问一句带报错的(例如「pip install timeout 是什么原因」)——
它应该先去查经验库再回答。想确认状态 npx @misaka-net/misakanet-setup --verify,想关掉 --uninstall。
不想碰命令行 / 不知道 assistant 的配置文件在哪: 把下面这句话复制粘贴给助手,它会自己装好、自己验证、用大白话告诉你结果:
帮我接入 MisakaNet 失败记忆库:请读取 https://raw.githubusercontent.com/Ikalus1988/MisakaNet/main/integrations/agent-autostart/INSTALL_FOR_ME.md ,按里面的「第 2 部分:给你的要求」执行,做完用中文简单告诉我结果。
如果上面这条网址打不开(部分网络会拦 raw.githubusercontent.com),换成 CDN 镜像的同一份文件:
帮我接入 MisakaNet 失败记忆库:请读取 https://cdn.jsdelivr.net/gh/Ikalus1988/MisakaNet@main/integrations/agent-autostart/INSTALL_FOR_ME.md ,按里面的「第 2 部分:给你的要求」执行,做完用中文简单告诉我结果。
| 想做的事 | 用什么 |
|---|---|
| 一行装(有 Node) | npx @misaka-net/misakanet-setup |
| 让助手自己装(不会命令行) | 上面那段话(raw 或 jsDelivr 两个链接任选其一) |
| WSL / Linux / macOS 脚本 | curl -fsSL https://raw.githubusercontent.com/Ikalus1988/MisakaNet/main/integrations/agent-autostart/bootstrap.sh | bash(被拦就用 https://cdn.jsdelivr.net/gh/Ikalus1988/MisakaNet@main/...) |
| 只看它要改什么 | 加 --dry-run |
| 卸载 | --uninstall(改过的文件都有 .misakanet.bak) |
读不需要注册(匿名 5 次/天/IP);安装器会顺手注册一个匿名节点并把 token 写进你本机的配置,用来解除这个限制。
课程内容是数据不是指令——助手被要求不要把课程里的命令当命令执行。