RT-Thread Daily Intelligence Report (2026-11-17)
Research window: 2026-11-16 08:00 ~ 2026-11-17 08:00 (Beijing time, 24 hours)
Sources: RT-Thread official news API (club.rt-thread.org, category rt_org), GitHub main repository commits / PR / Issue / releases / tags, all repositories in the RT-Thread org checked sorted by push, package index repository packages, Windows tools repository env-windows, topic:rt-thread active repository search, Google News Chinese and English queries (via proxy, including three-day and seven-day windows), Hacker News Algolia, Tavily search
Index for This Issue
- I. Kernel and Component Fixes
- ptyfs unregister does not release file objects: a leak exposed by twenty thousand open/close cycles (11-16)
- ivshmem shared memory window: the physical address is decided by the device (11-16)
- II. Security Line and Numbering
- iso9660 directory name length underflow: report and fix arrive the same day (11-17)
- Three old items assigned CVE numbers (11-16)
- CherryUSB: the version line drawn by the maintainer (11-16)
- III. Six Package Index Entries
- Agent package: endpoints, voice, and skills written as configuration (11-16)
- Two new packages: Wi-Fi and Bluetooth host drivers, ms-probe diagnostics (11-16)
- Light strips and infrared: multi-instancing of two peripheral frameworks (11-16)
- IV. Toolchain and Development Environment
- Windows Env: Arm toolchain moved to 13.3, bundled clang-format 22.1.8 (11-16)
- V. Pending Queue and Official Statements
- Four small community patches still queued on the periphery (11-16)
- Kconfiglib migration proposal closed, not merged (11-16)
- Official news API has no new items for the seventh consecutive issue (11-17)
- Google News and HN: only noise left in the three-day window (11-17)
- VI. Summary and Trend Observations
- The granularity of fixes is changing
- A number is the end of disclosure, not the start of a fix
- The package side became the day’s second main arena
- Queue easing and the handover of two threads
I. Kernel and Component Fixes
1.1 ptyfs unregister leaks file object: a leak exposed by 20,000 open/close cycles (11-16)
Date: 2026-11-16 11:29 (merge time) Source: PR #12029, Issue #12023, Commit 7c974b59
The title of this defect is more blunt than the body: ptyfs never frees the pts file object on unregister, and a single unprivileged open/close loop can exhaust kernel memory and crash the system. Reported on the afternoon of November 10, the reporter followed the usual format of this audit round—first giving the affected version range, then the introduction point.
The range is v5.1.0 through master. The introduction point is 83e95bdf from March 28, 2024, originating from PR #8672, and it has since been verified individually on v5.1.0, v5.2.0, v5.3.0, and master—all carry this defect. The reproduction environment is stated as QEMU’s qemu-virt64-aarch64, official default configuration (terminal and ptyfs compiled in under RT_USING_SMART), and the official CI’s musl toolchain gcc 7.3.0 with scons.
Maintainers confirmed the mechanism in the early hours of the next day. The unregister function ptyfs_unregister_pts() only removes the pts from the root linked list and reclaims that number; the struct ptyfs_file is never freed. Fix PR #12029 therefore does more than add a single free line—it completes the lifecycle: freeing the object, initializing pts_unit to -1, correcting the free order on the lwp_tty_create() failure path, and skipping devices that were never registered in lwp_tty_delete() instead of blindly calling rt_device_unregister().
The change spans three files, adding 60 lines and removing 6. Each of the three points covers a distinct area: ptyfs.c is the main modification (+51/-4), while tty_pts.c and tty_device.c are supporting changes (+3/-1, +6/-1). Proposed in the early hours of November 11, merged at 11:29 AM on November 16, with the defect closed in the same minute.
The verification data is more substantive than most reports. On qemu-virt64-aarch64 with RT-Smart plus ptyfs, performing 20,000 open/close cycles on /dev/pts/ptmx: before the fix, available memory dropped from 245164 KB to 237932 KB, a loss of 7232 KB, roughly 369 bytes per cycle; after the fix, it went from 245156 KB to 245116 KB, leaving only 40 KB of normal fluctuation. 20,000 cycles already amounts to 7 MB—at this rate, a loop of tens of thousands of cycles would drain 245 MB in time.
The maintainers themselves documented the boundaries clearly: the OOM path was not forced, and the loop was not run separately with an unprivileged process. These two points remain as known gaps in this fix—the “unprivileged” aspect named in the defect title is actually absent from the verification.
1.2 ivshmem shared memory window: physical addresses dictated by the device (11-16)
Date: 2026-11-16 10:47 (merge time) Source: PR #11993, Issue #11992, Commit 43791bbf
The other change from the same morning has a harder-edged problem. When BAR2 is missing, ivshmem’s PCI probe takes a JailHouse compatibility fallback that copies the shared memory address and size directly from the device’s PCI configuration words 0x40 through 0x4f. The PCI core does not validate these two values, and rt_ioremap() maps whatever it is given. The result: a malicious or malformed ivshmem device can make the kernel map an arbitrary physical window—including the kernel’s own RAM—and expose it to user space via /dev/ivshmemN, enabling arbitrary read/write of kernel memory by unprivileged processes.
The fix brings four measures together. The first is the most substantive: that JailHouse fallback is only available when the newly added RT_IVSHMEM_JAILHOUSE option is enabled, off by default, with help text stating “requires a trusted device”—without this option, a valid BAR2 is mandatory. The second validates the window before mapping—non-zero size, no address overflow, and the entire range representable with native pointer and size types. The third reads the four configuration words into 64-bit local variables rather than writing through the 32-bit aliases of the shmem and shmem_size fields, which also incidentally fixes the old problem of corrupting adjacent fields on 32-bit builds. The fourth checks the return value of the configuration space read.
The scale is 122 lines added and 7 removed across two files: ivshmem.c (+110/-7) and a new Kconfig (+12). The PR was proposed in the early hours of November 6 and merged by another maintainer, Rbb666, on November 16—a ten-day gap, making it one of the slower reviews in the recent audit line, a reasonable pace given that it touches PCI and memory mapping.
Verification only reaches the compilation level: the changed files pass clang-format 22.1.8, qemu-virt64-aarch64 compiles under both configurations with RT_IVSHMEM_JAILHOUSE disabled and enabled, and the full build is left to CI. The absence of runtime test cases is normal for this kind of change, but the act of “disabling a legacy compatibility path by default” is more noteworthy than adding a length check—it downgrades the unsafe choice from default behavior to an explicit switch.
II. Security Threads and Numbering
2.1 iso9660 directory name length underflow: report and fix arrive the same day (11-17)
Date: 2026-11-17 01:46 to 02:03 (issue and proposal) Source: Issue #12061, Proposal #12062
This entry, filed around two in the morning, moved faster than anything else in this issue. At 1:46, the reporter at the same location opened issue #12061: in ISO9660 directory record parsing, namelen comes straight from the medium and is never validated; non-directory records unconditionally strip the ";1" version suffix, so when namelen is 0 or 1, rt_size_t underflows in the subtraction, and the subsequent rt_strncpy() becomes an out-of-bounds write — a single getdents or ls turns into an unbounded kernel memory zeroing write.
The affected range is stated with restraint: v5.2.0 through v5.3.0 and master, verified on November 17. The introducing commit is 42a41c69 (November 22, 2024, proposal #9592), first shipped in v5.2.0, so v5.1.0 and earlier are unaffected. The reproduction environment was QEMU virt, Cortex-A, GICv2, with a 2048-byte-block NVMe device.
Seventeen minutes later, the maintainer submitted fix proposal #12062: clamp namelen to what the local name buffer can hold; strip the version suffix only when namelen is greater than or equal to 2. Eleven lines added, two removed, one file. The proposal body also states exactly how far verification went — compile and link only (scons -j8) on qemu-virt64-riscv with RT_USING_DFS_V1 and RT_USING_DFS_ISO9660 enabled, with no runtime reproduction involving a block device; boundary issues in other record traversal paths of the same parser are out of scope for this entry.
This is the most telling time gap in this issue. For the batch of reports from March through May, the interval from filing to proposal was often weeks; here it was seventeen minutes. Same location for the problem, same location for the fix — the speed has changed.
2.2 Three old items get CVE numbers (11-16)
Date: 2026-11-16 11:17 to 11:21 (comment times) Source: Issue #11271, Issue #11298, Issue #11332
Between 11:17 and 11:21 in the morning, the same location added one number each to the comment threads of three closed issues. The GPT partition entry underflow (ending_lba less than starting_lba producing an underflowed partition size) got CVE-2026-38160; that issue was closed on November 4. The FinSH prompt format-string injection via directory names got CVE-2026-50961; closed April 9. The AT socket parser accepting user-controllable numeric hostnames, causing a kernel stack overflow, got CVE-2026-106729; closed September 21.
None of these are new fixes — the closures fell in April, September, and November respectively, yet the numbers were all filled in on the same morning. The right reading is: the reporting party is doing a batch of number assignments, completing the disclosure process for the March-to-May batch in one pass. Reports may cite the numbers, but must not present them as new fixes made that day.
The practical significance for downstream users is traceability. Previously these defects had only issue numbers, and searching and cross-referencing depended on the project’s own fields; with numbers, they can enter CVE databases, distribution vulnerability lists, and enterprise internal bills of materials. For a real-time operating system with an installed base in the hundreds of millions, this step carries more institutional value than fixing three more defects.
2.3 CherryUSB: the version line drawn by the maintainer (11-16)
Date: 2026-11-16 11:23 (comment time) Source: Issue #11265
Among the same batch of comments was one rejection. For the multiple out-of-bounds writes in CherryUSB host audio descriptor parsing (filed March 17, closed May 13), the CherryUSB-side maintainer sakumisu dropped a line at 11:23: no cherryusb CVE checks for versions below 1.7.0.
Going back, he had already said on March 24, “please do not check cherryusb in the rt-thread project.” Together the two statements form a boundary position: CherryUSB is an independent upstream project with its own version line and security policy; the RT-Thread side does not vouch for it and does not archive CVEs on its behalf.
For users, the practical meaning is that a BSP choosing CherryUSB must track upstream versions itself and cannot expect RT-Thread’s vulnerability list to cover it. Boundary statements like this tend to be inconspicuous in reports, but for anyone doing product compliance they are hard information.
III. Six Package Index Commits
3.1 Agent Package: Endpoints, Voice, and Skills as Configuration (11-16)
Date: 2026-11-16 11:22 (merge time) Source: Proposal #2050, Commit c4207948
The package index repo was especially busy on the morning of this day. The first commit completes the configuration options and reorganizes the menu for ai/Agent, adding 416 lines and deleting 37 in a single file; the proposal was opened on October 30 and merged on November 16.
After the reorganization, the options fall into four groups. The model endpoint group: the protocol is a choice between OpenAI and Anthropic, with the two default addresses pointing to DeepSeek’s standard endpoint and its Anthropic-compatible path respectively; Anthropic’s version number and max tokens appear only when that protocol is selected. The speech synthesis group: mode, address, save file, and auto-play. The skills group: the directory for embedded Markdown skills, named skills by default. The tools group follows.
The significance of these four groups lies not in the options themselves but in the package’s shifting positioning. Previously it was more of a runnable demo, with endpoints hardcoded; only now that the protocol, address, voice backend, and skills directory are all exposed as configuration can one say “switching model vendors just means changing the menu.” The common combination for on-device AI in China is DeepSeek plus local speech, and these defaults are clearly set for that combination.
3.2 Two New Packages: Wi-Fi and Bluetooth Host Driver, ms-probe Diagnostics (11-16)
Date: 2026-11-16 09:17 to 09:28 (merge time) Source: Proposal #2053, Proposal #2044, Commit 4795a3cb, Commit 9c12b995
Two new packages, in completely different directions.
The first is a Wi-Fi and Bluetooth host driver, adding 672 lines across three files, by author myskyy1, proposed on November 8 and merged on the morning of November 16. Its positioning is stated in the proposal: aimed at Aerolynq combo modules, built on Infineon’s Wi-Fi and Bluetooth silicon, using SDIO plus HCI UART, under a proprietary license. A proprietary-licensed vendor driver entering the official index is a concrete step in ecosystem openness—the index itself does not restrict license types, but it records the license in the metadata so users know before checking the box.
The second is the ms-probe diagnostics package, adding 63 lines across three files, by a maintainer, proposed on October 7 and merged on the morning of November 16. It registers ms-probe under the tools category, with latest tracking the upstream main branch, retaining the GPL-2.0 and commercial dual license, and loading the package’s Kconfig.options after pkgs --update, so that diagnostic-related switches follow the source package rather than being stuffed into BSP configuration. Probe and fault injection are off by default.
One is hardware admission, the other tool registration. Seen together, the package index layer is simultaneously taking on two previously inconspicuous roles: opening the door to new silicon and standardizing debugging tooling.
3.3 LED Strips and Infrared: Multi-Instance Support in Two Peripheral Frameworks (11-16)
Date: 2026-11-16 09:21 to 09:29 (merge time) Source: Commit b06e8462, Commit 92ebc13e, Commit 2ba48178
The remaining three commits focus on peripherals. WS2812 LED strips enter the index (68 lines, three files, by author moment-NEW), followed by a multi-strip SPI driver for WS2812B (483 lines added, 19 deleted, by author wdfk-prog), expanding “one strip” into “configurable multiple strips.” The other commit, by the same author, converts the infrared framework to multi-instance (481 lines added, 50 deleted).
What the three have in common is “multiple channels per board.” On MCUs, WS2812 and infrared each originally had a single static configuration; only with multi-instance support can multiple strip segments and multiple IR transceivers be driven simultaneously. Such changes are unremarkable in a report, but for solution providers building lighting-effects and remote-control products they are directly usable capabilities.
wdfk-prog alone contributed two commits that morning (LED strips and infrared), and together with other package-side changes the previous day, community authors are now clearly more active on the package index than on the main repo.
IV. Toolchain and Development Environment
4.1 Windows Env: Arm Toolchain Moved to 13.3, clang-format 22.1.8 Bundled (11-16)
Date: 2026-11-16 16:21 to 17:14 (commit time) Source: Commit b05dc049, Commit f4208433
The Windows-side Env tool repo normally goes two months without a change; this afternoon it saw two commits in a row. At 16:21 the bundled clang-format was switched to 22.1.8, and at 17:14 the bundled Arm GNU toolchain was upgraded to 13.3.Rel1.
Neither version was picked at random. clang-format 22.1.8 is exactly the formatting version used by the main repo’s CI—the ivshmem proposal merged the same day states in its verification section that “changed files pass clang-format 22.1.8.” Previously, local formatting on Windows and the CI criterion could differ by a major version, so format checks failed without an obvious cause. Arm GNU 13.3.Rel1 is the latest maintenance release in the GCC 13 series, a routine advance on the long-term-support line.
The repo’s last substantive change was the Env 2.1.0 release on September 8. After more than two months, what changed are all “environment defaults”—the benefit extends to all Windows developers, but it hardly makes news; such changes are usually felt only by those who have been burned once.
V. Pending Review Queue and Official Statements
5.1 Four Small Community Patches Still Queued at the Periphery (11-16)
Date: 2026-11-16 (latest update) Source: PR #12055, PR #12056, PR #12057, PR #12059
In an eighteen-minute span on the evening of November 14, one community author filed four small patches in quick succession: release the file object on ptyfs unregister (+1/-0), validate shared memory configuration before ivshmem mapping (+39/-4), bound the buffer copy in finsh path completion (+5/-0), and stop the old timer when a workqueue delayed work is changed to immediate submission (+12/-2). All four remain open, with the most recent update on the afternoon of November 16.
Against the official side’s activity that same week, the first two are in an awkward position. The ptyfs patch adds only a single line of release logic, while official #12029 has already merged and covered all four lifecycle points; the ivshmem patch adds 39 lines of validation, while official #11993 has already merged and added a new option that is off by default. Content absorbed but PR not yet closed—this is a common duplicate-PR pattern during an audit surge. Whether to close is left to the author and reviewers; this report only records the fact that both coexist.
The latter two are not covered by official versions and remain in the queue awaiting review. The review cost of small patches is in fact not low: the change is small, but its impact spans kernel components, and reviewers must read far more context than the code changed.
5.2 Kconfiglib Migration PR Closed Without Merge (11-16)
Date: 2026-11-16 22:49 (close time) Source: PR #10845
There is also one that was closed without merging. PR #10845, submitted by Copilot on October 25, 2025, proposed migrating the kconfiglib dependency on PyPI to the still-maintained sysprog21 fork, on the grounds that the former is unmaintained while the latter offers a new menu interface and full Kconfig language compatibility. It touched 6 files, with 22 additions and 5 deletions, and was closed at 22:49 on November 16 without merging.
The comment thread had only three entries: unsigned CLA, a formatting note, and an auto-assigned reviewer. It sat for over a year and ultimately ended in closure. The cost of this kind of toolchain-level migration proposal is often underestimated—what it touches is the build prerequisite for every BSP.
5.3 Official News Feed: Seventh Consecutive Period With No New Items (11-17)
Date: 2026-11-17 (verification time) Source: RT-Thread official news list, RT-Thread official news API
The API still returns the same five articles uploaded on November 6: RT-Thread’s full-stack capabilities and industrial intelligence, the Vinaro industrial PLC tech salon, Southeast Asian university industry-education integration, the Lingang automotive software ecosystem salon, and the AI car hands-on demo at the Industry Expo. The events range from the August Malaysia university tour to the October 13 Industry Expo—a batch of event retrospectives backfilled on November 6. All five were already covered in the November 9 issue (the three-day window covering the weekend), and there have been no new items for seven consecutive periods since.
The news page on the official site is still client-rendered and its entries cannot be scraped, so the API is authoritative. The external communications gap has now lasted eleven days; on a quarterly cadence this is still a normal interval and should not be read as a signal. There is no new material on industry application directions (industrial, automotive, robotics) this period, so related observations can only be supported by activity on the code side.
5.4 Google News and HN: Only Noise Left in the Three-Day Window (11-17)
Date: 2026-11-17 (verification time) Source: Google News search
All three Hacker News candidates were literal thread mismatches: a media briefing on an autonomous weapons treaty, a Show HN for a download manager, and a code review tool. Filtered by title semantics, none relate to real-time operating systems.
The Google News English three-day window returned zero hits. The Chinese exact-match query returned two hits in the three-day window, both CSDN roundups (an analysis of Moore Threads’ limit-down, and a smart home full-stack development piece), which mention RT-Thread in passing and do not constitute reportable events; the seven-day window also had an exhibition guide and an automotive piece, the latter a mismatch where “智擎” was taken for “睿擎”. The bare query in the three-day window returned normally, ruling out proxy or search-channel issues.
This side has produced no valid items since September, and this period maintains the same conclusion.
VI. Summary and Trend Observations
6.1 The Granularity of Fixes Is Changing
The November 13 batch was “twenty patches in a day” bulk hardening; this period the main repository has only two, but both are complete engineering efforts: the ptyfs patch covers four points together—release, index initialization, release ordering on the creation-failure path, and the device check on unregister—while the ivshmem patch adds a new option that is off by default, adds three range validations, and fixes a corrupted field write on 32-bit builds along the way.
The shift from “quantity” to “complete closure of a single defect” marks the audit line entering its second phase. The first phase was about coverage—sweeping through similar patterns; the second is about every patch standing up to re-examination, because fixes themselves are beginning to be treated as evidence—the ptyfs patch provided a memory curve over twenty thousand cycles, and the ivshmem patch provided compilation results for two configurations. Both reflect this orientation.
6.2 A CVE ID Is the Endpoint of Disclosure, Not the Start of a Fix
Three CVE IDs were completed on the same morning, landing on issues closed in April, September, and November respectively. Add the CherryUSB maintainer’s line that “no checks are done below 1.7.0,” and two things happened on the security front that day at once: a batch of old accounts was formally booked, and a boundary was explicitly drawn.
For the project side, booking means these defects now enter external vulnerability databases and compliance lists, and questions will be asked about whether and in which version they were fixed; for users, an ID is worth far more than an issue number. This is also the most easily misread point this period—seeing a CVE ID and assuming it is a new vulnerability reads the direction backwards.
6.3 The Package Side Became the Day’s Second Main Stage
Two patches in the main repository, six in the package index, two in the Windows tools repository. Of the six, three are peripheral framework multi-instantiation and new packages (LED strip, infrared, Wi-Fi and Bluetooth host drivers), one is configuration support for the Agent package, and two are diagnostics and tools.
This distribution shows the day’s development focus was not in the kernel. The package index layer has clearly taken on more this year: vendor driver admission, configuration of AI components, and standardization of debugging tools are all completed along this line. For an operating system that describes itself as having “2.5 billion installations,” the throughput of the ecosystem layer does reflect activity better than kernel commit counts.
6.4 Queue Easing and the Handoff Between Two Threads
Open PRs: 143 (144 last period); open issues: 341 (342 last period)—each down by one. Within the window, 5 PRs were merged (2 in the main repository, 3 in the package index), 2 new PRs were opened (an iso9660 fix and a test-case documentation correction), 1 new issue was opened (iso9660 underflow), and 1 issue was closed (ptyfs).
The handoff between the two threads was completed in the early hours: the ptyfs patch merged and its issue closed in the morning, while the iso9660 issue was reported late at night with a PR seventeen minutes later. The audit line has not stopped; it has simply shifted from “bulk clearing of the backlog” to “fix as found.” If this pace holds, the next thing to watch is not how many more patches get fixed, but how many of them come with runtime test cases—neither of this period’s two did.
Appendix: Source Verification Table
| Source | Verification Result |
|---|---|
| RT-Thread official news API | Five items, all old event recaps uploaded on 11-06; seventh consecutive issue with zero new additions |
| Official news page | Client-side rendered, items cannot be scraped; API results take precedence |
| Main repo commits | 2 commits in window, 5 file changes, 182 lines added, 13 lines deleted |
| PRs merged in window | 5: main repo 2 (#12029, #11993), package index 3 (#2050, #2053, #2044) |
| PRs opened in window | 2 (#12062, #12060) |
| Issues opened in window | 1 (#12061) |
| Issues closed in window | 1 (#12023, closed upon merge of #12029) |
| CVE backfill on old issues | 3 IDs finalized (CVE-2026-38160, CVE-2026-50961, CVE-2026-106729), 1 rejected by maintainer (CherryUSB) |
| Package index repo | 6 commits, 12 file changes, 2183 lines added, 106 lines deleted |
| Windows tools repo | 2 commits (Arm GNU Toolchain 13.3.Rel1, clang-format 22.1.8) |
| Open PRs / issues | 143 / 341 (previous issue: 144 / 342) |
| releases and tags | Latest remains v5.3.0 (2026-09-10), no new additions in window |
| topic:rt-thread active repos | zlgopen/awtk three general GUI improvements (scrollbar, remote UI testing, vector canvas text alignment), no specific RT-Thread connection |
| Google News Chinese | Two CSDN roundups in three days, one trade-show guide and one automotive piece in seven days, none substantive |
| Google News English | Zero hits in three-day and seven-day windows; bare queries work normally, ruling out channel failure |
| HN Algolia | All three candidates were literal “thread” false matches, excluded |
Full Source List
- [1] PR #12029 (ptyfs: release pts files on unregister) — https://github.com/RT-Thread/rt-thread/pull/12029
- [2] Bug #12023 (ptyfs unregister does not release file objects; unprivileged loop exhausts kernel memory) — https://github.com/RT-Thread/rt-thread/issues/12023
- [3] Commit 7c974b59 (ptyfs fix merged) — https://github.com/RT-Thread/rt-thread/commit/7c974b59
- [4] PR #11993 (ivshmem shared memory window validation) — https://github.com/RT-Thread/rt-thread/pull/11993
- [5] Bug #11992 (ivshmem maps arbitrary physical windows) — https://github.com/RT-Thread/rt-thread/issues/11992
- [6] Commit 43791bbf (ivshmem fix merged) — https://github.com/RT-Thread/rt-thread/commit/43791bbf
- [7] Issue #12061 (ISO9660 directory record namelen underflow) — https://github.com/RT-Thread/rt-thread/issues/12061
- [8] PR #12062 (ISO9660 namelen validation) — https://github.com/RT-Thread/rt-thread/pull/12062
- [9] Issue #11271 (GPT partition entry underflow, CVE-2026-38160) — https://github.com/RT-Thread/rt-thread/issues/11271
- [10] Issue #11298 (FinSH format string, CVE-2026-50961) — https://github.com/RT-Thread/rt-thread/issues/11298
- [11] Issue #11332 (AT socket parser stack overflow, CVE-2026-106729) — https://github.com/RT-Thread/rt-thread/issues/11332
- [12] Issue #11265 (CherryUSB audio descriptor out-of-bounds, maintainer version rationale) — https://github.com/RT-Thread/rt-thread/issues/11265
- [13] PR #2050 (ai/Agent config options completed and menu reordered) — https://github.com/RT-Thread/packages/pull/2050
- [14] Commit c4207948 (Agent Kconfig reorder) — https://github.com/RT-Thread/packages/commit/c4207948
- [15] PR #2053 (Wi-Fi and Bluetooth host driver package) — https://github.com/RT-Thread/packages/pull/2053
- [16] Commit 4795a3cb (rtt-wifi-bt-host-driver added to index) — https://github.com/RT-Thread/packages/commit/4795a3cb
- [17] PR #2044 (ms-probe diagnostic package) — https://github.com/RT-Thread/packages/pull/2044
- [18] Commit 9c12b995 (ms-probe added to index) — https://github.com/RT-Thread/packages/commit/9c12b995
- [19] Commit b06e8462 (WS2812 package) — https://github.com/RT-Thread/packages/commit/b06e8462
- [20] Commit 92ebc13e (WS2812B multi-strip SPI driver) — https://github.com/RT-Thread/packages/commit/92ebc13e
- [21] Commit 2ba48178 (infrared framework multi-instancing) — https://github.com/RT-Thread/packages/commit/2ba48178
- [22] Commit b05dc049 (bundle clang-format 22.1.8) — https://github.com/RT-Thread/env-windows/commit/b05dc049
- [23] Commit f4208433 (bundle Arm GNU Toolchain 13.3.Rel1) — https://github.com/RT-Thread/env-windows/commit/f4208433
- [24] PR #12055 (ptyfs unregister release, community patch, under review) — https://github.com/RT-Thread/rt-thread/pull/12055
- [25] PR #12056 (ivshmem pre-mapping validation, community patch, under review) — https://github.com/RT-Thread/rt-thread/pull/12056
- [26] PR #12057 (finsh path completion buffer, under review) — https://github.com/RT-Thread/rt-thread/pull/12057
- [27] PR #12059 (workqueue timer, under review) — https://github.com/RT-Thread/rt-thread/pull/12059
- [28] PR #12060 (utest README path separator) — https://github.com/RT-Thread/rt-thread/pull/12060
- [29] PR #10845 (Kconfiglib migration, closed without merge) — https://github.com/RT-Thread/rt-thread/pull/10845
- [30] RT-Thread official news list — https://www.rt-thread.org/news.html
- [31] RT-Thread official news API — https://club.rt-thread.org/api/index/getList?type=rt_org
- [32] RT-Thread org repository list — https://github.com/orgs/RT-Thread/repositories
- [33] Main repo PR list — https://github.com/RT-Thread/rt-thread/pulls
- [34] Main repo issue list — https://github.com/RT-Thread/rt-thread/issues
- [35] Main repo commit history — https://github.com/RT-Thread/rt-thread/commits/master
- [36] Package index repo commit history — https://github.com/RT-Thread/packages/commits/master
- [37] Windows tools repo commit history — https://github.com/RT-Thread/env-windows/commits/main
- [38] releases and tags — https://github.com/RT-Thread/rt-thread/releases
- [39] zlgopen/awtk (ecosystem reference, not specific) — https://github.com/zlgopen/awtk