AI Stack Alert High CVE-2026-62331 For CISO, SOC / IR, AppSec, AI Eng AI-assisted
CVE-2026-62331: praktor Telegram upload path traversal enables persistent instruction injection
praktor, a multi-agent Claude Code orchestrator, prior to 0.9.1 concatenates the attacker-controlled Telegram document filename into destination paths in internal/telegram/bot.go without sanitization. path.Join collapses parent-directory traversal, so a remote Telegram user can write outside uploads/ within the persistent agent workspace at /workspace/agent. The issue is fixed in 0.9.1.
Why it matters
With allow_from: [] the bot accepts all senders, so no authentication is needed. An attacker can plant or overwrite CLAUDE.md, skill, hook, or configuration files that the agent reads on later runs, turning a file upload into persistent instruction injection.
Do this
- Upgrade praktor to version 0.9.1 or later.
- Set allow_from to an explicit list of trusted Telegram user IDs instead of an empty list.
- Inspect the workspace volume mounted at /workspace/agent for unexpected CLAUDE.md, skill, hook, or settings files changed outside uploads/.
- Compare agent instruction and configuration files against a known-good baseline and investigate unexplained changes before the next agent run.
Exploit Clock
Exploitation unlikely (for now)
- EPSS 0.78%
Signals
- NoCISA KEV
- 8.2CVSS
- 0.78%EPSS
- NoPublic PoC
- 36/100Priority
- 80%Confidence
Evidence
“Prior to 0.9.1, both Telegram upload sites in `internal/telegram/bot.go` concatenate the attacker-controlled document filename into a destination path without sanitization”
“Because `allow_from: []` permits all senders by default, no prior authentication or privilege is required in that configuration.”
“The attacker can plant or overwrite `CLAUDE.md`, instruction, skill, hook, or configuration files that the Claude agent reads and operates on across later runs”
“This issue is fixed in version 0.9.1.”
Sources
Affected / related
praktor, Telegram, path traversal, stored instruction injection, Claude Code
Published 2026-10-30 19:18 UTC · updated 2026-11-08 02:31 UTC · AI-assisted summary, checked against quoted sources · How we score