llamaCVEVulnerability Intelligence
/
Advisory · GitHub · Go · code.vikunja.io/veans

GHSA-CVH3-HF56-G9XM

Remote project and task metadata is injected into the coding agent's prompt

Sin parche publicado
Descargar PNG JSON
CVSS 3.1
3.7
Baja
CVSS:3.1/AV:N/AC:H/PR:L/UI:R/S:U/C:L/I:L/A:N Fuente: GitHub
ExplotaciónSin evidenciaNo figura en CISA KEV
Exploits públicos0Ninguno conocido
ParcheSin parcheEl fabricante no ha publicado corrección
Exposición28 dDesde la divulgación, 22 oct 2026, 5:00 a. m.
Detección27 dDesde la divulgación hasta su ingreso al índice
Timeline

Ciclo de vida de la vulnerabilidad

Hoy
Divulgación
Publicada
Explotación
Remediación
Monitoreo
Detectada +1
22 oct 26
18 nov 26
19 nov 26
22 oct 26
29 oct 26
5 nov 26
12 nov 26
19 nov 26
FechaFaseEventoDetalle
Divulgación Publicada GHSA-CVH3-HF56-G9XM divulgada · CVSS 3.7 (GitHub)
Monitoreo Detectada Ingresó al índice
Monitoreo Verificada Última verificación contra las fuentes
Resumen
Descripción del advisory
Texto original · GitHub · en inglés

Affects all veans versions (veans has no tagged release yet).

## Summary
A project member with write access sets a project title, or task titles and descriptions, that veans feeds verbatim into a coding agent's instructions on every session start. The text is rendered inside the authoritative instruction block, so the agent may follow it.

## Details
- `veans/internal/commands/prompt.tmpl:1-8` and `veans/internal/commands/prime.go:87-97` render the fields inside the authority block.
- The hook runs `prime` on SessionStart and PreCompact (`veans/internal/bootstrap/hooks.go:35,41`).
- Three of the four rendered fields come from `.veans.yml`, which adds nothing because the repo already controls CLAUDE.md. `ProjectTitle` (from the server) and task content from `list`/`show` are the real boundary.
- On an untrusted server (see the `VEANS_TOKEN` redirect advisory), the server itself controls all of this content.

## Impact
Prompt injection into an agent that can read and change the developer's checkout and run commands. Requires write access to the project (or a hostile server), and depends on the agent acting on the injected text.

## Fix
Two parts.
1. Trust gate: an untrusted server's metadata never reaches the prompt, because `prime` fails closed when it isn't interactive. It renders without remote metadata, plus a hint to run any veans command interactively to trust the server.
2. Fencing, which is still required: a writer on a *trusted* server can set `ProjectTitle` and task content. Render remote metadata and task content as a fenced, labeled untrusted-data block, outside the authority section.

The trust gate is a single origin check in the client/runtime setup (`veans/internal/commands/runtime.go:57-63`), with trust records in global XDG state keyed by normalized origin plus repo identity, an interactive confirmation, and fail-closed behavior when non-interactive.

_Found via internal security review (csf #3, 03 #13, deep #9)._

Vector CVSS 3.1

Desglose de métricas

Opción resaltada = valor del vector · izquierda = peor caso · GitHub
Vector de ataqueAV:N
RedAdyacenteLocalFísica
ComplejidadAC:H
BajaAlta
PrivilegiosPR:L
NingunoBajosAltos
InteracciónUI:R
NingunaRequerida
AlcanceS:U
CambiadoSin cambio
ConfidencialidadC:L
AltoBajoNinguno
IntegridadI:L
AltoBajoNinguno
DisponibilidadA:N
AltoBajoNinguno
Versiones

Rangos afectados

Afectada Corregida
code.vikunja.io/veans≤ 0.0.0
0.0.0
Remediación

Estado del parche por paquete

Paquete / productoEstadoCorregida enDesde divulgaciónÚltima versión
Gocode.vikunja.io/veans
Sin parche — — —
Explotación

Arsenal público

Sin exploits, módulos ni PoC públicos conocidos a la fecha de verificación.

Referencias
Advisoryhttps://github.com/go-vikunja/vikunja/security/advisories/GHSA-cvh3-hf56-g9xm

Las referencias externas se muestran para copiar: cve.llama.ec no lo redirige a otros sitios.