Résumé
skilo add follows symbolic links, allowing arbitrary local file disclosure from a malicious skill source
Détails de l’avis
Impact
skilo add installs a skill by recursively copying the skill directory into the
target skills directory. The copy routine (copy_dir_all) classified each entry
with std::fs::DirEntry::file_type() — which does not follow symlinks — and
then copied non-directory entries with std::fs::copy(), which does
dereference symlinks.
As a result, a skill containing a symbolic link such as
reference.txt -> /home/<user>/.ssh/id_rsa was copied as a regular file whose
contents are the link's target. A malicious skill source — for example a git
repository installed via skilo add github.com/<attacker>/<skills>, or a local
path — could read arbitrary files readable by the user running skilo add (SSH
keys, cloud credentials, .env files, etc.) and place their contents inside the
installed skill directory, where the user or their agent may later read, share,
or sync them.
This is arbitrary local file disclosure (CWE-59 / CWE-61, symlink following) triggered by installing an untrusted skills source.
Patches
Fixed in 0.11.1. copy_dir_all now rejects symbolic-link entries at any
recursion depth (failing closed with a dedicated error) instead of dereferencing
them.
Workarounds
- Only install skills from sources you trust.
- Inspect a skill source for symbolic links before running
skilo add.
Affected versions
Introduced together with the skilo add command in 0.5.0 and present through
0.11.0. Releases before 0.5.0 do not include the add command.
Références
- https://github.com/advisories/GHSA-6xx4-9wp6-65p7
- https://github.com/manuelmauro/skilo/security/advisories/GHSA-6xx4-9wp6-65p7
- https://github.com/manuelmauro/skilo/pull/11
- https://github.com/manuelmauro/skilo/commit/c14bdc2eddcf26633ab1dcc0b2d5c0ff42c72a3e
- https://github.com/manuelmauro/skilo/releases/tag/v0.11.1
Vulnérabilités liées
Tout Supply chain →- MEDIUMCVE-2026-81727
NLTK: Downloader.download follows hardlinks and overwrites outside-root files
- HIGHCVE-2026-75914
CodeWhale: image_analyze follows workspace symlinks, leaking external file bytes
- HIGHCVE-2026-81726
NLTK: Model-artifact APIs bypass pathsec and touch files outside allowed roots
- HIGHGHSA-2rx9-3g3h-c2jv
pnpm: pacquet trust-lockfile install can create dependency symlinks outside the project
- MEDIUMCVE-2026-55569
Aqua's archive extraction follows attacker-planted symlinks, allowing writes outside the install directory
- HIGHCVE-2026-55108
KubeVela Terraform remote loader DoS via unbounded file read