CVE Feed

Gitea's activity-feed and contribution-heatmap API handlers do not apply the profile visibility check for restricted accounts viewing Limited-visibility users. A restricted user can retrieve activity records and contribution timestamps that the corresponding profile routes deny. Exposed activity can include repository names, commit messages, and issue or pull request events associated with qualifying public-repository activity.

Gitea enforces first-time fork contributor approval by blocking jobs, but some job transition and dispatch paths do not re-check the run's approval requirement. A separate run sharing a job-level concurrency group can cancel a pending job and cause its blocked sibling jobs to be re-emitted and dispatched without approval. This allows unreviewed workflow code to execute on self-hosted runners; fork secrets remain withheld and the run token remains read-only.

For pull request events, Gitea reads workflow definitions from the pull request head and posts a synthetic skipped status when workflow branch or path filters exclude execution. A fork pull request author can supply a workflow matching a required status context but deliberately prevent it from running. Because skipped statuses count as success, branch protection or organization-required workflow checks can appear satisfied without the mandated workflow executing; no runner or maintainer approval is required for this path.

Gitea's user organization-list API includes Limited-visibility organizations for authenticated restricted accounts without applying the visibility restriction enforced by the single-organization endpoint. A restricted user can list a known user's public organization memberships and retrieve limited organizations' names, descriptions, and avatars. This exposes organization metadata that should be hidden from the requesting account.

Gitea's package registry grants authenticated users read access to packages owned by Limited-visibility users without excluding restricted accounts. A restricted user can enumerate and download those packages despite the visibility policy that should prevent this access. The issue affects user-owned package registries and exposes package contents across the intended authorization boundary.

The fix for CVE-2026-12996 was incomplete. A use-after-free in OpenVPN 2.6.0 through 2.6.22 and 2.7_alpha1 through 2.7.7 allows remote authenticated peers to potentially cause a denial of service or leak memory via crafted packets during TLS session promotion or expiry.

Gitea's collaborative-owner authorization path for reading cross-repository reusable workflows does not exclude fork pull request runs. An approved fork workflow targeting a private repository can read workflow files from another private repository that trusts the base repository's owner through collaborative-owner configuration. The issue exposes private reusable workflow contents at caller-selected Git refs; it does not itself grant write access or code execution in the other repository.

Gitea's Swift package parser limits each manifest to 128 KiB but does not limit the number of manifests or their aggregate size before reading them into memory. An authenticated user with package-publish permission can upload a highly compressed archive containing many manifests, causing allocations far larger than the uploaded archive. This can exhaust memory, terminate the process, or cause request failures.

Gitea parses Alpine package uploads before enforcing package-size quotas and does not bound the decompressed .PKGINFO data or its dependency and provides entry counts. An authenticated user with package-publish permission can upload a small, highly compressed archive that causes disproportionately large memory allocations during parsing and metadata generation. On memory-constrained deployments, this can terminate the process or otherwise disrupt availability.

Gitea's initial GitLab migration version probe uses an HTTP client without a timeout and does not attach the migration task's cancellation context. An authenticated user who can start migrations can target an attacker-controlled server that keeps the response open indefinitely. The request continues after task cancellation and can exhaust the migration worker pool, preventing other migrations from progressing.

Gitea's OneDev migration downloader reads the entire remote version response into memory without a size limit and uses a client without a timeout. An authenticated user who can start a repository migration can direct it to an attacker-controlled OneDev-compatible server returning an oversized or never-ending response. This can consume memory and hold migration workers, impairing service availability.

Gitea's repository-side team-link API allows a repository administrator who is not an organization owner to attach organization teams to a repository despite RepoAdminChangeTeamAccess being disabled. The endpoint does not enforce the organization's restriction on repository administrators changing team access. This allows unauthorized changes to which organization teams receive repository access.

Gitea's Jupyter notebook renderer inserts the notebook's language_info.name metadata into a CSS class attribute while disabling sanitization and rendering directly in the main document without an iframe sandbox. A malicious repository notebook can supply multiple CSS classes that create a viewport-covering overlay and intercept clicks when another user views it. The reported impact is UI redress; script execution was not demonstrated.

Gitea's Org-mode markup renderer processes INCLUDE directives supplied to the repository markup endpoint without preventing access to server-side files. An unauthenticated attacker can use a suitable public repository to read files accessible to the Gitea service account. If the disclosed files include app.ini and its INTERNAL_TOKEN, the attacker can chain the disclosure with internal logger access to modify a Git hook and execute commands as the Gitea service account.

Gitea's starred-repository and tracked-time API endpoints do not re-check current repository permissions before returning data associated with the requesting user. A user whose access to a private repository has been revoked can still retrieve repository metadata through /api/v1/user/starred and private issue titles through /api/v1/user/times. The disclosure depends on relationships or tracked-time entries created before access was revoked.

When OpenID sign-in is enabled, Gitea accepts a user-supplied OpenID URL and performs discovery requests without validating the destination against internal network addresses. An unauthenticated remote attacker can cause the server to contact loopback or private-network HTTP services. This can expose internal service behavior and enable interaction with resources that are not directly reachable by the attacker.

The thumbnail Python package through 1.5 contains an OS command injection vulnerability in generate_thumbnail() that allows attackers to execute commands by supplying shell metacharacters in file paths. Attackers who control a file name passed to generate_thumbnail, such as an uploaded file, can inject commands into os.system() calls running ffmpeg, convert or unoconv with the Python process privileges.

Stills before 25.24.2 contains an OS command injection vulnerability that allows attackers to execute shell commands by supplying crafted media file paths. Attackers can name a media file with $(...) or backticks to execute operating system commands with Node.js process privileges when ffmpeg, ffprobe, gifsicle, or convert runs.

@offactory/compress-fonts before 1.1.1 contains an OS command injection vulnerability that allows attackers to execute shell commands by supplying font files with crafted names. Font paths are interpolated unquoted into a FontForge command run through child_process.execSync, so shell metacharacters in file names execute commands with the Node.js process privileges.

Turnpress through 0.4.2 contains an OS command injection vulnerability that allows attackers to execute shell commands by supplying .docx files with crafted names. The convertDocxToHtml function in src/converter.ts invokes pandoc through execa with shell:true, so shell metacharacters in the --docx path run with the victim's privileges during convert or create.

Beganovich snappdf from 1.1.0 before 6.1.0 contains an OS command injection vulnerability on Windows that allows attackers to run commands by influencing the rendered URL. Attackers can supply cmd.exe metacharacters such as & in the setUrl() value, which executeOnWindows() passes unescaped to PHP exec(), executing commands as the PHP process.

Video-stitch before 2.0.0-beta.1 contains an OS command injection vulnerability that allows attackers to execute commands by supplying file names interpolated unquoted into shell-executed ffmpeg command lines. Attackers who influence file names passed to concat(), cut(), or merge() can inject shell metacharacters to run commands with the Node.js process privileges.

nodejs-whisper through 0.3.1 contains an OS command injection vulnerability in convertToWavType() that allows attackers to execute commands by supplying crafted audio file paths. Attackers can provide file names containing $() command substitution or closing quotes, which are interpolated into an ffmpeg command run via shelljs exec(), executing OS commands with Node.js process privileges.

transcribe2texts through 1.0.6 contains an OS command injection vulnerability that allows attackers to execute commands by supplying crafted audioPath values to transcribeAudio(). Attackers can inject $() or backtick substitution or a closing quote into the unescaped path passed to child_process.exec() to run commands with Node.js process privileges.

xhtml2pdf before 0.2.19 contains an OS command injection vulnerability in pisa.startViewer() that allows attackers to execute commands by interpolating unescaped file names into os.system. Attackers can supply an input file whose name contains a double quote and shell metacharacters, which runs OS commands as the invoking user when the pisa tool is used with -s/--start-viewer.

@runnerty/executor-decompress through 3.1.1 contains an OS command injection vulnerability that allows attackers to execute commands by supplying crafted archive file names to compress_file. Attackers who can name a file dropped into a watched directory can inject $() substitutions or unquoted shell metacharacters into shelljs exec() calls, running commands with Runnerty process privileges.

Whisper-node through 1.1.1 contains an OS command injection vulnerability that allows attackers to execute commands by supplying crafted filePath or whisperOptions.language values to whisper(). Attackers can inject shell metacharacters or $() substitution into the command string built by createCppCommand() and run via shelljs exec() to execute commands with Node.js process privileges.

@mborne/extract before 0.2.1 contains an OS command injection vulnerability that allows attackers to execute commands by supplying shell metacharacters in archivePath or targetDir options. Attackers controlling these values passed to extract() can inject commands into unquoted shelljs exec() strings in the unzip, 7z, tar-bz2 and gz extractors, running them with Node.js process privileges.

doc-recog through 1.1.6 contains an OS command injection vulnerability that allows attackers to execute commands by supplying crafted document paths to recog() or tag(). Attackers can embed $(...) or backtick command substitution in the file path, which DocData._readPDF passes to a shell-based pdftotext exec call, running commands with Node.js process privileges.

buildebooks before 0.2.2 contains an OS command injection vulnerability in lib/kindle.js that allows attackers to execute commands by supplying a crafted EPUB file name passed to kindlegen via child_process.exec. Attackers can craft a book project with a malicious front-matter filename or project folder name and kindle: true to execute OS commands as the user running buildebooks.

pypdftk through 0.5 contains an OS command injection vulnerability in fill_form(), dump_data_fields(), and pdftk_cmd_util() that builds shell commands from caller-supplied paths. Attackers controlling pdf_path or out_file can inject shell metacharacters interpreted by subprocess.Popen with shell=True to execute OS commands as the Python process.

webp-converter before 3.0.0 contains an OS command injection vulnerability in src/webpconverter.js helpers that run libwebp binaries via execFile with shell enabled. Attackers controlling file names or option strings passed to cwebp, dwebp, gwebp, or webpmux functions can inject shell metacharacters to execute OS commands as the Node.js process.

git-diff before 2.0.7 contains an OS command injection vulnerability that allows attackers to execute shell commands by supplying crafted strings to the gitDiff API. Attackers who control a diffed document string can embed $(cmd) or backtick substitutions that shelljs.exec evaluates, running commands with Node.js process privileges.

MyReport Updater through 2.1.1 contains a local privilege escalation vulnerability that allows low-privileged users to make the SYSTEM updater service install attacker-supplied archives into arbitrary directories. Attackers can use UpdaterManager.exe over gRPC to set a malicious reference server and installLocation, planting a DLL like HID.dll loaded by osk.exe to gain SYSTEM privileges.

Report One MyReport Updater through 2.1.1 exposes an unauthenticated gRPC service on TCP port 9095 that allows remote attackers to redirect the SYSTEM updater by invoking the UpdateConfigReferent method. Attackers can point the updater to a malicious reference server serving modified module archives with matching MD5 digests to write files and execute code as SYSTEM.

Komodo through 2.3.3 contains an authentication bypass vulnerability in the Periphery agent that allows unauthenticated attackers to connect as a trusted Core because empty core_public_keys and passkeys lists fail open by default. Attackers reaching port 8120 can present any generated keypair to issue Periphery API requests, including terminal sessions and container operations, executing commands on managed hosts typically as root.

VentureDrake Laravel CRM before 2.4.0 contains an incorrect authorization vulnerability in user management that allows authenticated administrators to assign unrestricted roles via Role::find. Attackers with user-management access can assign the Owner role or other teams' roles and delete cross-team users through the unscoped UserIndex delete() method.

VentureDrake Laravel CRM before 2.4.0 contains a missing authorization vulnerability that allows authenticated users to bypass Spatie role permissions through Livewire action methods. Read-only users can invoke delete, won, lost or reopen methods on components like DealIndex, or use unguarded Activities and Products routes, to modify deals, quotes, orders and invoices.

Kan 0.5.5 through 0.6.0 contains an unauthenticated server-side request forgery vulnerability in the /api/download/attatchment route due to an incomplete fix for CVE-2026-32255. When S3_ENDPOINT is unset, attackers can pass arbitrary url query parameters to fetch() to read internal services, localhost ports, and cloud instance metadata.

ClipBucket v5 through 5.5.3-#206 contains a reflected cross-site scripting vulnerability in admin_area/index.php that allows attackers to inject HTML via the val parameter in mode=add_todo. Attackers can lure a logged-in administrator into submitting a cross-site POST form without CSRF protection, executing script in the admin panel origin.

ClipBucket v5 through 5.5.3-#206 contains a reflected cross-site scripting vulnerability in admin_area/language_settings.php via the unencoded edit_language GET parameter. Attackers can send administrators with basic_settings permission a crafted link whose payload is echoed into a hidden input, executing script in the admin panel origin.

ClipBucket v5 through 5.5.3-#206 contains a reflected cross-site scripting vulnerability in admin_area/flagged_item.php via the unencoded type GET parameter. Attackers can send crafted flagged_item.php?type= links to logged-in administrators or moderators to execute injected script in the admin panel origin.

Papermerge Core through 3.5.3 contains a missing authorization vulnerability that allows authenticated users to list other groups' tags and document types by supplying a group_id parameter. Attackers who know a group's UUID can query GET /tags/all or GET /document-types/all, which skip the user_belongs_to check, to disclose metadata of groups they do not belong to.

Papermerge Core through 3.5.3 contains a missing authorization vulnerability in the shared nodes API that allows authenticated users to share other users' documents by supplying arbitrary node UUIDs. Attackers can call POST /shared-nodes or PATCH /shared-nodes/access/{node_id} to grant themselves any role, then read, modify, or delete victims' documents and revoke existing shares.

Papermerge Core through 3.5.3 contains a missing authorization vulnerability that allows authenticated users to manipulate other users' pages because page management and OCR endpoints skip has_node_perm checks. Attackers knowing page UUIDs can call /pages/, /pages/move, or /pages/extract to rotate, delete, or extract victims' pages, and trigger OCR via /tasks/ocr.

Swetrix Cloud before 5.1.0 contains a missing authorization vulnerability in ProjectController.addProject that allows authenticated users to hijack arbitrary projects by moving them into their own organisation. Attackers can create an organisation and add a victim project by its public tracking ID to read analytics, change settings, or delete it.

Formbricks before 4.9.0 contains an authorization bypass vulnerability in server actions like resetSurveyAction and updateSegmentAction that fail to verify resource ownership against the supplied organizationId. Authenticated owners or managers of any organization can pair their own organization identifier with victim resource IDs to delete survey responses, modify segments and quotas, or alter pending invitations.

frp 0.41.0 through 0.71.1 contains a missing authentication vulnerability that allows unauthenticated remote attackers to access pprof debug endpoints because handlers are registered on the root mux outside Basic Authentication. Attackers can query /debug/pprof/ endpoints on frps dashboard or frpc admin listeners to obtain cmdline, goroutine stacks, heap profiles, and trigger CPU profiling.

SmsForwarder through 3.5.0 contains a missing authentication vulnerability in its AndServer HTTP server that allows unauthenticated network attackers to invoke remote-control APIs. Attackers can call the default-enabled endpoints to read SMS, call logs and contacts, send SMS, and export or overwrite configuration via the clone API.

we-mp-rss through 1.5.3 contains a missing authorization vulnerability in the /api/v1/wx/tools/export/download endpoint that allows unauthenticated attackers to download export archives because its authentication dependency is commented out. Attackers can supply an mp_id and guessable timestamp-based export filename to retrieve archives under data/docs, and set delete_after_download=true to delete them.