Skip to content
Agent Threat Rules (ATR)中文
ΕTool Poisoning
ATR-2026-01963
Recorded 2026-06-29 · Author ATR Community

PraisonAI Action Orchestrator step.target Path Traversal Arbitrary File Write RCE (CVE-2026-39305 / GHSA-jfxc-v5g9-38xr)

Detects CVE-2026-39305 (GHSA-jfxc-v5g9-38xr, CWE-22, CRITICAL): PraisonAI (< 4.5.113) Action Orchestrator builds a write path as `workspace / step.target` without resolving or boundary-checking it.

path traversal arbitrary file write · pattern · confidence: high

Severity
critical
Lanes
alert · huntmarks and reports a match
Maturity
testnot yet settled
Scan target
both
Status
experimental

Framework mappings

OWASP Agentic Top 10
  • ASI05:2026
  • ASI02:2026
OWASP LLM Top 10
  • LLM06:2025

ATR wrote these mappings. The framework owners have not reviewed or endorsed them.

Description

Detects CVE-2026-39305 (GHSA-jfxc-v5g9-38xr, CWE-22, CRITICAL): PraisonAI (< 4.5.113) Action Orchestrator builds a write path as `workspace / step.target` without resolving or boundary-checking it. An ActionStep of type FILE_CREATE / FILE_EDIT whose `target` contains `../` traversal escapes the workspace and writes arbitrary files (e.g. ~/.ssh/authorized_keys, ~/.bashrc), yielding RCE. This rule keys on the Action Orchestrator sink tokens (ActionStep / step.target / FILE_CREATE / FILE_EDIT) co-occurring with relative traversal sequences and sensitive write targets.

Suggested responsealertblock input
Author: ATR Community · Created 2026-06-29 · Last modified 2026-11-18

Detection conditions

Combinator: any
  1. 01
    one line: PraisonAI ActionStep with a FILE_CREATE/FILE_EDIT target value containing 3+ ../ traversal segments — the core step.target escape
    field: contentop: regex
  2. 02
    one line: relative traversal chain writing to the advisory PoC sink (orchestrator_pwned) or sensitive RCE targets (authorized_keys/.bashrc/.profile)
    field: contentop: regex
  3. 03
    one line: PraisonAI / Action Orchestrator context combined with traversal sequences and the workspace/step.target write sink
    field: contentop: regex
  4. 04
    explicit CVE framing
    field: contentop: regex

Attacks it catches

  1. ActionStep(id='x', action_type=ActionType.FILE_CREATE, target='../../../../../../../tmp/orchestrator_pwned.txt', params={'content':'pwned'})
  2. PraisonAI Action Orchestrator step.target = '../../../../../../.ssh/authorized_keys' via workspace / step.target FILE_CREATE write
  3. CVE-2026-39305: PraisonAI Action Orchestrator path traversal in step.target allows arbitrary file write (authorized_keys) leading to RCE

Attack examples the rule must catch, many based on reported attacks and some written to match their shape. They are versioned with the rule as regression tests. If a later revision stops catching one, it shows.

Benign inputs it leaves alone

  1. Legitimate Action Orchestrator FILE_CREATE writing a normal relative path inside the workspace — no ../ traversal
    ActionStep(action_type=ActionType.FILE_CREATE, target='reports/summary.md', params={'content':'done'})
  2. Defensive mention of authorized_keys with no traversal sequence and no orchestrator/ActionStep sink
    Incident response checklist: review ~/.ssh/authorized_keys for SSH backdoors and rotate keys
  3. Mitigation advice citing the CVE without any traversal payload or exploit tokens
    Upgrade PraisonAI to 4.5.113 to remediate CVE-2026-39305

Known false positives

  • ▸Legitimate PraisonAI ActionStep FILE_CREATE/FILE_EDIT whose target is a normal relative filename inside the workspace (no ../ traversal).
  • ▸Incident-response or hardening checklists that mention ~/.ssh/authorized_keys or .bashrc defensively, without any path-traversal write.
  • ▸Security advisory text quoting CVE-2026-39305 as a fixed/mitigated issue without an executable traversal payload.
title: "PraisonAI Action Orchestrator step.target Path Traversal Arbitrary File Write RCE (CVE-2026-39305 / GHSA-jfxc-v5g9-38xr)"
id: ATR-2026-01963
rule_version: 1
status: experimental
description: >
  Detects CVE-2026-39305 (GHSA-jfxc-v5g9-38xr, CWE-22, CRITICAL): PraisonAI
  (< 4.5.113) Action Orchestrator builds a write path as `workspace / step.target`
  without resolving or boundary-checking it. An ActionStep of type FILE_CREATE /
  FILE_EDIT whose `target` contains `../` traversal escapes the workspace and writes
  arbitrary files (e.g. ~/.ssh/authorized_keys, ~/.bashrc), yielding RCE. This rule
  keys on the Action Orchestrator sink tokens (ActionStep / step.target / FILE_CREATE /
  FILE_EDIT) co-occurring with relative traversal sequences and sensitive write targets.
author: "ATR Community"
date: "2026/06/29"
schema_version: "0.1"
detection_tier: pattern
maturity: test
severity: critical
references:
  owasp_llm: ["LLM06:2025"]
  owasp_agentic: ["ASI05:2026", "ASI02:2026"]
  mitre_atlas: ["AML.T0053 - AI Agent Tool Invocation"]
  mitre_attack: ["T1190 - Exploit Public-Facing Application"]
  cve: ["CVE-2026-39305"]
metadata_provenance: { mitre_atlas: human-reviewed, owasp_llm: human-reviewed, owasp_agentic: human-reviewed }
compliance:
  eu_ai_act:
    - article: "15"
      context: "Article 15 (accuracy, robustness, cybersecurity) — runtime detection of this technique is a cybersecurity control for high-risk AI systems. Technique: PraisonAI Action Orchestrator step.target Path Traversal Arbitrary File Write RCE (CVE-2026-39305 / GHSA-jfxc-v5g9-38xr)."
      strength: primary
    - article: "9"
      context: "Article 9 (risk management system) requires identified risks to be addressed by appropriate measures; this rule is a runtime risk-treatment control. Technique: PraisonAI Action Orchestrator step.target Path Traversal Arbitrary File Write RCE (CVE-2026-39305 / GHSA-jfxc-v5g9-38xr)."
      strength: secondary
  nist_ai_rmf:
    - subcategory: "MP.5.1"
      context: "NIST AI RMF MAP 5.1 — likelihood and impact of the identified attack are characterised; this rule detects the adversarial input at runtime. Technique: PraisonAI Action Orchestrator step.target Path Traversal Arbitrary File Write RCE (CVE-2026-39305 / GHSA-jfxc-v5g9-38xr)."
      strength: primary
    - subcategory: "MG.3.2"
      context: "NIST AI RMF MANAGE 3.2 — runtime monitoring/maintenance control that surfaces this attack class. Technique: PraisonAI Action Orchestrator step.target Path Traversal Arbitrary File Write RCE (CVE-2026-39305 / GHSA-jfxc-v5g9-38xr)."
      strength: secondary
Show all (58 more lines)
  iso_42001:
    - clause: "8.1"
      context: "ISO/IEC 42001 Clause 8.1 (operational planning and control) — detection of this payload is an operational control. Technique: PraisonAI Action Orchestrator step.target Path Traversal Arbitrary File Write RCE (CVE-2026-39305 / GHSA-jfxc-v5g9-38xr)."
      strength: primary
    - clause: "8.3"
      context: "ISO/IEC 42001 Clause 8.3 (AI risk treatment) — this rule implements runtime detection as a treatment control. Technique: PraisonAI Action Orchestrator step.target Path Traversal Arbitrary File Write RCE (CVE-2026-39305 / GHSA-jfxc-v5g9-38xr)."
      strength: secondary
tags: { category: tool-poisoning, subcategory: path-traversal-arbitrary-file-write, scan_target: both, confidence: high, source: cve-disclosure, vendor_sources: praisonai-cve-2026-39305 }
agent_source: { type: llm_io, framework: [praisonai, any], provider: [any] }
detection:
  condition: any
  false_positives:
    - "Legitimate PraisonAI ActionStep FILE_CREATE/FILE_EDIT whose target is a normal relative filename inside the workspace (no ../ traversal)."
    - "Incident-response or hardening checklists that mention ~/.ssh/authorized_keys or .bashrc defensively, without any path-traversal write."
    - "Security advisory text quoting CVE-2026-39305 as a fixed/mitigated issue without an executable traversal payload."
  conditions:
    - field: content
      operator: regex
      value: '(?i)(?:ActionStep|action[_-]?type|step\.target|target\s*=)[\s\S]{0,200}(?:FILE_CREATE|FILE_EDIT)?[\s\S]{0,120}target\s*[=:]\s*["'']?(?:\.\.[\\/]){3,}'
      description: "one line: PraisonAI ActionStep with a FILE_CREATE/FILE_EDIT target value containing 3+ ../ traversal segments — the core step.target escape"
    - field: content
      operator: regex
      value: '(?i)(?:\.\.[\\/]){3,}[^\n"'']{0,120}(?:tmp/orchestrator_pwned|\.ssh/authorized_keys|/\.bashrc|/\.bash_profile|/\.profile)'
      description: "one line: relative traversal chain writing to the advisory PoC sink (orchestrator_pwned) or sensitive RCE targets (authorized_keys/.bashrc/.profile)"
    - field: content
      operator: regex
      value: '(?i)(?:praisonai|action[_-]?orchestrator)[\s\S]{0,200}(?:\.\.[\\/]){2,}[\s\S]{0,80}(?:FILE_CREATE|FILE_EDIT|step\.target|workspace\s*/\s*step)'
      description: "one line: PraisonAI / Action Orchestrator context combined with traversal sequences and the workspace/step.target write sink"
    - field: content
      operator: regex
      value: '(?i)CVE-2026-39305[\s\S]{0,200}(?:path[\s_-]?traversal|step\.target|action[\s_-]?orchestrator|FILE_CREATE|authorized_keys|arbitrary[\s_-]?file)'
      description: "explicit CVE framing"
response:
  actions: [alert, block_input]
  message_template: >
    [ATR-2026-01963] CRITICAL: PraisonAI Action Orchestrator step.target path traversal (CVE-2026-39305) writes arbitrary files outside the workspace (e.g. ~/.ssh/authorized_keys) enabling RCE. Block this ActionStep, canonicalise/validate target paths against the workspace boundary, and upgrade PraisonAI to >= 4.5.113.
confidence: 80
test_cases:
  true_positives:
    - input: "ActionStep(id='x', action_type=ActionType.FILE_CREATE, target='../../../../../../../tmp/orchestrator_pwned.txt', params={'content':'pwned'})"
      expected: triggered
      description: "Verbatim advisory PoC: FILE_CREATE ActionStep with deep ../ traversal target to /tmp/orchestrator_pwned.txt"
    - input: "PraisonAI Action Orchestrator step.target = '../../../../../../.ssh/authorized_keys' via workspace / step.target FILE_CREATE write"
      expected: triggered
      description: "Action Orchestrator traversal writing the SSH backdoor sink described in the advisory"
    - input: "CVE-2026-39305: PraisonAI Action Orchestrator path traversal in step.target allows arbitrary file write (authorized_keys) leading to RCE"
      expected: triggered
      description: "Explicit CVE-2026-39305 reference with path-traversal and authorized_keys exploit context"
  true_negatives:
    - input: "ActionStep(action_type=ActionType.FILE_CREATE, target='reports/summary.md', params={'content':'done'})"
      expected: not_triggered
      description: "Legitimate Action Orchestrator FILE_CREATE writing a normal relative path inside the workspace — no ../ traversal"
    - input: "Incident response checklist: review ~/.ssh/authorized_keys for SSH backdoors and rotate keys"
      expected: not_triggered
      description: "Defensive mention of authorized_keys with no traversal sequence and no orchestrator/ActionStep sink"
    - input: "Upgrade PraisonAI to 4.5.113 to remediate CVE-2026-39305"
      expected: not_triggered
      description: "Mitigation advice citing the CVE without any traversal payload or exploit tokens"

Revision history

Created
2026-06-29
Last modified
2026-11-18
Full commit history on GitHub