Skip to content
KitploitKITPLOIT
ToolsExploitsBlog
Log in
Submit
ToolsExploitsBlog
Submit

Hacking, PenTest, and Cybersecurity Tools for Your Security Arsenal!

Kitploit is a directory of hacking, cybersecurity, and pentesting tools. Discover the latest project updates to find vulnerabilities, analyze systems, automate testing, and strengthen your security.

FeedsContactPrivacy© 2026 Kitploit

Tool Directory

Categories

View all categories
Loading categories
loom-ci-name-adopt — A Loom agent admin can name a harness so it adopts another group's code interpreter, then delete that interpreter while cleaning up their own agent. They still cannot read the other group's agent. v1.10.3. High 7.6. | Kitploit
Tools/GitHubGitHub/abraxas/loom-ci-name-adopt
Privilege EscalationVulnerability AnalysisExploitationCloud SecurityIdentity & Access Management (IAM)Labs & Practice
GitHubabraxas/loom-ci-name-adopt

loom-ci-name-adopt

Share

A Loom agent admin can name a harness so it adopts another group's code interpreter, then delete that interpreter while cleaning up their own agent. They still cannot read the other group's agent. v1.10.3. High 7.6.

View Repository
1 day agoNot yet reviewed

Abraxas Labs - loom-ci-name-adopt

loom-ci-name-adopt

An agent administrator in one group can name a harness so the code-interpreter name Loom derives is the name another group already has. The new row stores that existing interpreter id. A later delete of the administrator's own agent, with cloud cleanup on, deletes that interpreter.

They still cannot open the other group's agent. The lab did not run code as the other group's role. AWS was stubbed. What was recorded is the id being stored, a different name not taking that id, and cleanup deleting the stored id.

Class: Account-wide name collision on a cloud resource, adopt on create and delete on cleanup Reach: Remote. In-group agent:write (g-admins-demo). They supply an execution role and a code-interpreter role that already belong to their own group.

Loom v1.10.3 cc10ab0. This container ran on that tag. I did not run this oracle on v1.9.0.

IDno CVE yet
ClassHarness code-interpreter name is not tied to the agent or the group
ReachRemote, agent:write in the caller's own group
CWECWE-863
CVSSHigh: 7.6 CVSS:3.1/AV:N/AC:L/PR:H/UI:N/S:C/C:N/I:L/A:H
Productawslabs/loom
Affectedv1.10.3 cc10ab08573087fa12154f7c8181ffcd985fea88
Authg-admins-demo. GET of the payments agent stayed 403.
LicenseGNU Affero GPL v3.0
Labloopback Docker, AWS clients stubbed, stack removed after the run

Confidentiality is unscored because the other group's agent row was not readable. Availability is the interpreter delete.

What an attacker can do

PrivilegeMethod / sinkImpactLab-proven or inferred
Be refused the other group's agentGET /api/agents/{billing}ClosedLab-proven 403. Body did not contain the interpreter id.
Store the existing interpreter id on a new harnessPOST /api/agents name billing_code_interpreter, code interpreter enabled, own rolesIntegrity. Both names strip to loom_ci_billing.Lab-proven. Id ci-billing-existing stored, deployment READY, runtime id set.
Create a distinct interpreter for a different base namePOST name payrollNegative controlLab-proven. Stored ci-created-1, created name loom_ci_payroll.
Delete the adopted interpreterDELETE /api/agents/{own}?cleanup_aws=true once runtime_id is setAvailabilityLab-proven. deleted-ids=ci-billing-existing
Run as the other group's interpreter rolen/aNot claimedNot labbed. The stub recorded ids.
Assign loom:group=payments or bind that group's rolecreate/delete guardsClosedDemo can only create and delete in demo.

Stays closed

AxisMethod / sinkWhy it is closed
ReconGET or list of the payments agent as demo403
IntegrityA harness whose stripped name differsDoes not reuse the billing id
AvailabilityDELETE without cleanup_aws, or with an empty runtime idCleanup returns before the interpreter delete
IntegrityManaged config keys and foreign IAM rolesSeparate checks. This bug does not reopen them.

list_code_interpreters is not paginated. A large account can miss the summary, and then this adopt does not see the target.

One footgun in the deploy function, not the bug: memory_records is only assigned when memory_ids is sent, and a later branch reads it when a code-interpreter role is set. The lab sends memory_ids: [0] so that name exists. An unknown id yields an empty list.

How I found it

Harness names have to be unique as strings. The interpreter name is loom_ci_ plus the harness name with a trailing code_interpreter stripped, then cut at 48 characters. billing and billing_code_interpreter are different rows and the same interpreter name. The create path lists interpreters in the account and, on a name match, saves that id on the new agent. There is no group compare.

I ran this on v1.10.3 cc10ab0. The stub listed loom_ci_billing as ci-billing-existing, recorded creates, and recorded deletes. The product source in the container was not patched.

Lab

python3 loom-ci-name-adopt-Abraxas-Labs.py
SUCCESS LOOM-1103-CI-ADOPT adopted=ci-billing-existing payroll-distinct=yes cleanup-deleted=yes LOOM-1103-CI-ADOPT-WITNESS

run.sh checks out v1.10.3, checks cc10ab0, and tears the stack down.

The fix

Put the agent id, or the loom:group, in the interpreter name, and refuse to store an id whose name was not created for this row. On cleanup, delete only an interpreter this agent created. Page the list.

References

  • awslabs/loom v1.10.3 cc10ab08573087fa12154f7c8181ffcd985fea88
  • agents.py _deploy_harness_background
  • name derivation
  • CWE-863
  • CVSS 3.1 calculator

License

GNU Affero GPL v3.0. See LICENSE. Loopback lab only. No warranty.

Abraxas Labs [email protected]

Download Tool

Most Popular

View all →

Discover the most used tools by our community.

No popularity data yet.

Explore all tools

Browse our collection of tools

View all tools →