CHECKPOINT

Human approves $500. Agent executes $50,000. The checkpoint is the exploit.
7
Subsystems
291
Tests
2
Critical CVEs
57
NIGHTFALL Tool
checkpoint-tool survey --target <URL>
LangGraph TOCTOU human-approval bypass CVE-2026-28277 / SQLite msgpack deserialization RCE CVE-2025-64439 / Cross-tenant thread_id enumeration / S3 checkpoint replacement mid-execution / Redis LPUSH checkpoint injection / Human sees transfer — agent executes different amount / Time-travel replay attacks / State persistence across security boundaries / Checkpoint stores hold full agent context LangGraph TOCTOU human-approval bypass CVE-2026-28277 / SQLite msgpack deserialization RCE CVE-2025-64439 / Cross-tenant thread_id enumeration / S3 checkpoint replacement mid-execution / Redis LPUSH checkpoint injection / Human sees transfer — agent executes different amount / Time-travel replay attacks / State persistence across security boundaries / Checkpoint stores hold full agent context

Agent Checkpoints Are Trusted Unconditionally — They Shouldn't Be

LangGraph, AutoGen, and every other agent framework that supports human-in-the-loop control relies on checkpointing: save state, wait for approval, resume. The assumption is that the state you saved is the state you resume. That assumption is wrong. CHECKPOINT exploits the gap between human approval and agent execution — across SQLite, Redis, S3, and in-memory stores.

LangGraph TOCTOU Human-Approval Bypass

LangGraph interrupt() saves state then waits for human approval. Between save and resume, an attacker modifies the checkpoint. Human approved $500 transfer — agent executes $50,000. The approval is detached from execution.

CVE-2026-28277 TOCTOU LANGGRAPH

SQLite msgpack Deserialization RCE

LangGraph SQLite checkpointer deserialises msgpack payloads without schema validation. Crafted checkpoint triggers arbitrary code execution on the agent host at resume time. Full RCE from a checkpoint file.

CVE-2025-64439 DESERIALIZATION RCE

Cross-Tenant Thread ID Enumeration

LangGraph thread_id values are sequential integers. An attacker with one valid thread_id enumerates adjacent threads, accessing other users' agent state including conversation history, credentials cached in context, and pending tool calls.

IDOR THREAD ENUM DATA EXFIL

S3 Checkpoint Replacement

Cloud-hosted agents checkpoint to S3. S3 bucket misconfiguration (public write, overly permissive IAM) allows an attacker to replace a checkpoint mid-execution. Agent resumes from attacker-controlled state.

S3 MID-EXECUTION STATE REPLACEMENT

Redis LPUSH Injection

Redis-backed checkpoint stores accept arbitrary writes if ACLs are misconfigured. CHECKPOINT injects malicious state entries that the agent loads at next resume, redirecting tool calls and goal state.

REDIS LPUSH GOAL HIJACK

Time-Travel Replay

Checkpointed state snapshots can be replayed to rewind agent execution to a prior decision point. Attacker replays from a checkpoint preceding a security check that has since passed, re-executing privileged operations.

REPLAY TIME-TRAVEL PRIVILEGE REPLAY

From Enumeration to Persistent Implant

Seven subsystems cover every phase of checkpoint exploitation: passive enumeration of the attack surface, TOCTOU injection through the approval window, deserialization RCE via crafted msgpack payloads, time-travel replay, cross-tenant state extraction, persistent goal-drift implants, and signed WARLORD-compatible reporting.

01
SURVEY
Enumerates checkpoint stores — SQLite databases, Redis instances, S3 buckets, and in-memory stores. Maps thread IDs, checkpoint timestamps, agent identities, and state schemas. Produces prioritised attack surface for subsequent subsystems.
PASSIVE — ANALYSIS
02
INJECT
Exploits the LangGraph TOCTOU window (CVE-2026-28277). Modifies checkpoint state between human interrupt() and agent resume. Substitutes tool call parameters, goal directives, and approval values. The human approved a checkpoint the agent never sees.
UNLEASHED --override
03
SURGERY
Weaponises CVE-2025-64439. Crafts malicious msgpack payloads that execute arbitrary code on the agent host when the SQLite checkpointer deserialises at resume. Full RCE from a checkpoint file.
UNLEASHED --override
04
REPLAY
Captures valid checkpoint state and replays it to rewind agent execution to a prior decision point. Bypasses security checks that passed after the captured checkpoint. Re-executes privileged operations from a clean state the agent trusts.
UNLEASHED --override
05
CROSS
Enumerates adjacent thread_id values from a single known ID. Extracts state from co-tenant agent sessions — conversation history, cached credentials, pending tool calls, and goal state of other users' agents.
PASSIVE + ACTIVE
06
PERSIST
Establishes persistent presence in the checkpoint store. Injects state entries that survive agent restart. Embeds goal-drift directives that accumulate over multiple checkpoints, gradually redirecting agent behaviour.
UNLEASHED --override --confirm-destroy
07
REPORT
Ed25519-signed, SHA-256-hashed reports. JSON (WARLORD-compatible) and Markdown. CVE mapping, thread enumeration results, injection parameters, and WARLORD handoff receipt.
ALL MODES

Standard Scan. Then Unleashed.

Standard mode — SURVEY + CROSS (read-only) + REPORT:

$ checkpoint-tool survey --target https://agent.example.com
[SURVEY] Enumerating checkpoint stores...
  SQLite: agent.db47 checkpoints found
  Thread IDs: sequential integers detected (100–146)THREAD-ENUM-001
  Redis: :6379ACL misconfiguration — write permitted
  S3: agent-checkpoints-prodpublic write detected
[CROSS] Enumerating adjacent threads from ID 100...
  41 adjacent threads accessible — conversation history, cached credentials extracted
[REPORT] Generating signed report...
  Findings: 3 CRITICAL, 2 HIGH, 1 MEDIUM
  CVEs: CVE-2026-28277, CVE-2025-64439
SCAN COMPLETE | Signed ✓ | WARLORD-compatible JSON output
  JSON: reports/checkpoint-survey-2026-04-24.json
  MD: reports/checkpoint-survey-2026-04-24.md

UNLEASHED mode — all attack subsystems. Requires Ed25519 key + signed scope:

$ checkpoint-tool inject --target https://agent.example.com --thread-id 142 --override

TOCTOU Exploitation

Targets the race window between LangGraph interrupt() and resume(). Substitutes approval values, tool parameters, and goal directives inside the window.

Deserialization RCE

Crafts malicious msgpack payloads exploiting CVE-2025-64439. Arbitrary code execution on the agent host at checkpoint resume — no prior access required.

Ed25519 Signed Reports

Every report cryptographically signed with Ed25519. SHA-256 evidence chains. WARLORD-compatible JSON output. Tamper-evident by design.

WARLORD Integration

CHECKPOINT is WARLORD-registered. Findings feed into autonomous campaign orchestration. Every checkpoint vulnerability becomes a campaign pivot point.

Multi-Store Coverage

SQLite, Redis, S3, in-memory — CHECKPOINT targets every checkpoint backend in production use. One tool, every store type, every known attack vector.

7
Subsystems
291
Tests
CVE-2026-28277
TOCTOU
CVE-2025-64439
RCE
57
Tool Number

Known Vulnerabilities CHECKPOINT Exploits

CHECKPOINT maps each subsystem to specific CVEs and vulnerability classes. Every finding in a CHECKPOINT report includes the relevant identifier, subsystem responsible, and a technical description of the exploitation mechanism.

Identifier Description Subsystem Notes
CVE-2026-28277 LangGraph interrupt() TOCTOU human-approval bypass INJECT Checkpoint modified between approval and resume
CVE-2025-64439 LangGraph SQLite checkpointer msgpack deserialization RCE SURGERY Arbitrary code execution at agent resume
THREAD-ENUM-001 LangGraph sequential thread_id IDOR CROSS Cross-tenant state extraction
S3-CHECKPOINT S3 checkpoint store IAM misconfiguration SURVEY / INJECT Mid-execution state replacement

Attack Surface, Standards, and Output

4 Store Types

Checkpoint Backends

  • SQLite (LangGraph default checkpointer)
  • Redis (distributed agent deployments)
  • Amazon S3 (cloud-native agent infra)
  • In-memory (development / test environments)
  • PostgreSQL (LangGraph production checkpointer)
  • Custom serialisation formats
Cryptographic

Report Integrity

  • Ed25519 digital signatures
  • SHA-256 evidence chains
  • WARLORD-compatible JSON output
  • Tamper-evident by design
  • Markdown + JSON dual output
  • Thread enumeration results preserved
OWASP LLM

Standards Mapping

  • LLM06 — Excessive Agency
  • LLM07 — System Prompt Leakage
  • LLM08 — Vector & Embedding Weaknesses
  • LLM02 — Sensitive Information Disclosure
  • LLM04 — Data and Model Poisoning
  • OWASP API Security — BOLA/IDOR

CHECKPOINT in the NIGHTFALL Kill Chain

CHECKPOINT is Tool 57 in the NIGHTFALL offensive pipeline — 65 tools across every layer. It operates at the agent execution layer, targeting the state persistence infrastructure that underpins human-in-the-loop workflows. Findings feed directly into WARLORD for autonomous campaign orchestration.

Tool 48
CRUCIBLE
Agent framework exploitation
Tool 49
VANTAGE
Agent telemetry & log injection
Tool 50
CIPHER
Cryptographic attack & disruption
Tool 55
FOUNDRY
Inference server exploitation
Tool 56
ADAPTER
LoRA/PEFT supply chain attack
Tool 57 — Active
CHECKPOINT
Agent state persistence exploitation
Tool 58
DELEGATE
Agent identity & OAuth delegation
Tool 59
PHANTOM SKILL
AI agent supply chain attack
Orchestration
WARLORD
Autonomous campaign orchestration
Defence
AI Shield
Defend everything above it

Security Distros & Package Managers

Kali Linux
.deb package
Parrot OS
.deb package
BlackArch
PKGBUILD
PyPI
pip install
macOS
pip install
Windows
pip install
Docker
docker pull
RS OS
Pre-installed
Pure Engineering
Zero Wrappers. Every Attack Vector Written From Scratch.

CHECKPOINT does not rely on existing exploitation frameworks. Every TOCTOU injection routine, every msgpack payload generator, every Redis LPUSH attack vector, every S3 replacement sequence — written from scratch in pure Python. No subprocess calls. No external tool dependencies. Real exploitation, not orchestrated scripts.

7
Subsystems
291
Tests Passing
2
Critical CVEs
0
External Dependencies
Ed25519 Cryptographic Override
CHECKPOINT UNLEASHED

Standard mode runs SURVEY + CROSS + REPORT — read-only enumeration with signed findings. UNLEASHED activates INJECT, SURGERY, REPLAY, and PERSIST — the destructive attack subsystems. Requires a valid Ed25519 private key and signed scope authorisation. One operator. Founder's machine only.

Standard
checkpoint-tool survey --target <URL>
SURVEY + CROSS + REPORT (read-only)
Override
checkpoint-tool inject --target <URL> --thread-id <ID> --override
All attack subsystems. Requires Ed25519 key + signed scope.

Authorised Use Only

Red Specter CHECKPOINT is intended for authorised security testing only. Exploitation of agent checkpoint stores on systems you do not own or have explicit written permission to test may violate the Computer Misuse Act 1990 (UK), Computer Fraud and Abuse Act (US), and equivalent legislation in other jurisdictions. CHECKPOINT is a professional offensive security tool for use by qualified security researchers and penetration testers against authorised targets only. Apache License 2.0.