kindling microVM protection
Reviews kindling's `kling` CLI (Firecracker/Virtualization.framework microVMs with a docker-style interface) and the `kling-db`, `kling-mcp`
Reviews kindling's `kling` CLI (Firecracker/Virtualization.framework microVMs with a docker-style interface) and the `kling-db`, `kling-mcp` and `kling-phone` extension binaries: credential and secret handoff and database copies at the strictest tier, removal and rollback of machines, volumes, templates and databases, provisioning of new machines, MCP services and extensions, and everyday command execution inside an existing machine. Read-only commands (ps, logs, inspect, top, status, doctor, version, machine audit, graph audit, db audit, the ls/inspect subcommands) match no rule here; kling is not on Guard's built-in safe-command list, so the unknown-executable review floor still applies to them. No safe variants are offered: kling answers -h/--help anywhere before `--` without running the command, but an environment variable kling reads can switch that interception off and a declarative matcher cannot see the inherited environment.
command.kindling · v1.0.0 · Other extensions
- Protection rules
- 5
- Permission checks
- 5
- Mapped commands
- 5
Overview
What this coverage does
Reviews kindling's `kling` CLI (Firecracker/Virtualization.framework microVMs with a docker-style interface) and the `kling-db`, `kling-mcp` and `kling-phone` extension binaries: credential and secret handoff and database copies at the strictest tier, removal and rollback of machines, volumes, templates and databases, provisioning of new machines, MCP services and extensions, and everyday command execution inside an existing machine. Read-only commands (ps, logs, inspect, top, status, doctor, version, machine audit, graph audit, db audit, the ls/inspect subcommands) match no rule here; kling is not on Guard's built-in safe-command list, so the unknown-executable review floor still applies to them. No safe variants are offered: kling answers -h/--help anywhere before `--` without running the command, but an environment variable kling reads can switch that interception off and a declarative matcher cannot see the inherited environment.
Key facts
- Catalog ID
command.kindling- Version
- v1.0.0
- Kind
- Command coverage
- Category
- Other extensions
- Protection rules
- 5
- Permission checks
- 5
- Mapped commands
- 5
- Maintainer
- Community contributor(community, provenance recorded)
- Source commit
- e8a9bf3
- Protection model
- External · opt-in
External coverage. It must be enabled through Guard controls; this page does not activate it.
Stated limits
- Coverage is limited to the reviewed operations and the surrounding Guard policy.
- A maintainer profile is not a security certification or an official upstream endorsement.
Command mapping
Every command this extension recognizes, with the catalog default floor Guard applies before workspace policy. Search the table, then open an operation for safe variants and the permission ID.
5 operations · 2 reapproval by default · 3 reviewed by default These are catalog defaults, not your workspace policy.
Showing 5 of 5 operations
- Reapproval Guard requires a fresh approval each time this command matches.Critical
kling machine credential api -domain api.example.com -env API_KEY
kindling credential or secret handoff
- Default floor
- Requires reapproval by default. Guard requires a fresh approval each time this command matches.
- Detector mode
- review
- What it matches
- Identifies commands that move a real secret into or out of a kindling microVM. `kling machine credential` and `kling template credential` hand an API key or a Postgres/MySQL password to kindling's credential proxy (read from a file or stdin, tied to one machine or to every machine later born from a template); `kling machine secret` (legacy `kling mmds`, `kling machine mmds`) injects a session sec…
- Permission ID
- command.kindling.permission.credential-handoff
- Reapproval Guard requires a fresh approval each time this command matches.Critical
kling db fork orders -n 3
kindling database copy or share
- Default floor
- Requires reapproval by default. Guard requires a fresh approval each time this command matches.
- Detector mode
- review
- What it matches
- Identifies `kling db` commands that duplicate or share database contents: `db fork` (N live copies of a running copy, data included), `db branch` (a copy per git branch forked from its parent), `db clone` and `db slice` (read a production Postgres URL and dump it, masked, into a microVM), and `db attach` (gives an agent in another machine access to a copy through the proxy). The same subcommands…
- Permission ID
- command.kindling.permission.data-copy
- Reviewed Guard intercepts the command for review before it runs.High
kling run -image python -egress internet
kindling machine, service or extension provisioning
- Default floor
- Reviewed by default. Guard intercepts the command for review before it runs.
- Detector mode
- review
- What it matches
- Identifies commands that start new code or bring new code in: `kling run` and `kling try` (a microVM running an image, optionally with -egress internet and -share SRC:DST:rw, which mounts a host folder writable), `sandbox create`/`new` and `sandbox fork` (N copies of a live sandbox, memory and processes included), `graph up` and `graph fork`, `save` (legacy `commit`), `image build`, `image put`,…
- Permission ID
- command.kindling.permission.provisioning
- Reviewed Guard intercepts the command for review before it runs.High
kling volume rm data
kindling machine, volume, template or database removal
- Default floor
- Reviewed by default. Guard intercepts the command for review before it runs.
- Detector mode
- review
- What it matches
- Identifies commands that destroy or roll back kindling state: `kling rm` and `kling stop` (machine and its snapshot), `sandbox rm`, `graph rm`, `volume rm` and `volume restore` (overwrites the current volume, keeping only `<name>@undo`), `template rm` (legacy `rmi`, `snapshots rm`), `image rm`, `plugin rm`/`uninstall`, and in the extensions `db rm`, `db reset`, `db undo`, `db env down`, `db class…
- Permission ID
- command.kindling.permission.resource-removal
- Reviewed Guard intercepts the command for review before it runs.Medium
kling exec sb-1 -- python main.py
kindling command execution inside an existing machine
- Default floor
- Reviewed by default. Guard intercepts the command for review before it runs.
- Detector mode
- review
- What it matches
- Identifies `kling exec`, `kling shell` and `kling cp`, which run a command inside, or copy a file into or out of, a microVM that already exists. This is everyday use: the machine's network egress and shares were fixed when it was created (and are reviewed by the provisioning rule), and kindling only accepts exec/cp on machines started with -allow-exec or created as sandboxes. It is listed at medi…
- Permission ID
- command.kindling.permission.guest-exec
Tool state mapping
No per-tool overrides are declared. Command defaults above resolve through the workspace Guard policy unless a later policy layer changes them.
Runtime identity
- Catalog ID
- command.kindling
- Guard enforcement ID
- command.kindling
- Source path
- contributions/extensions/command.kindling.json
- Contribution digest
sha256…44ea6d
Action classes
kindling credential handoffkindling database copykindling resource removalkindling machine provisioningkindling guest command executionTechnical profile
No independent profile published
This extension has no current independent technical profile. Catalog facts and any legacy launch remain separate and are not presented as profile evidence.
FAQ
Frequently asked questions
Reviews kindling's `kling` CLI (Firecracker/Virtualization.framework microVMs with a docker-style interface) and the `kling-db`, `kling-mcp` and `kling-phone` extension binaries: credential and secret handoff and database copies at the strictest tier, removal and rollback of machines, volumes, templates and databases, provisioning of new machines, MCP services and extensions, and everyday command execution inside an existing machine. Read-only commands (ps, logs, inspect, top, status, doctor, version, machine audit, graph audit, db audit, the ls/inspect subcommands) match no rule here; kling is not on Guard's built-in safe-command list, so the unknown-executable review floor still applies to them. No safe variants are offered: kling answers -h/--help anywhere before `--` without running the command, but an environment variable kling reads can switch that interception off and a declarative matcher cannot see the inherited environment. The listing declares 5 rules and 5 permission checks. 5 operations · 2 reapproval by default · 3 reviewed by default. Coverage is limited to these reviewed operations and the surrounding Guard policy.
External · opt-in: External coverage. It must be enabled through Guard controls; this page does not activate it. Enabling state is always controlled through Guard policy, never from this directory.
This listing entered the catalog as a community contribution through a public pull request with recorded provenance, with public credit to @juan52878911, and 1 GitHub maintainer is recorded on the listing. Credit is attribution only. A verified publisher profile and the claim flow verify authority separately; neither is an upstream endorsement or a HOL safety certification.
No. Every listing documents source, activation model, maintainer identity, and stated limitations so you can evaluate coverage before enabling it. Review the stated limitations and the exact source tree before relying on any single control.
5 operations · 2 reapproval by default · 3 reviewed by default. Examples: kling machine credential api -domain api.example.com -env API_KEY (requires reapproval by default); kling db fork orders -n 3 (requires reapproval by default); kling run -image python -egress internet (reviewed by default); kling volume rm data (reviewed by default); kling exec sb-1 -- python main.py (reviewed by default). These are catalog defaults, not your workspace policy.
External coverage. It must be enabled through Guard controls; this page does not activate it. Enablement is managed through your workspace's Guard policy and controls — never from this directory. Open the install guidance for the setup flow, then adjust the command coverage for command.kindling in Guard's policy surface.
The matching action is stopped at Guard's pre-action boundary before it executes, and the decision is recorded with evidence your workspace can review. Exact behavior follows your Guard policy combined with this entry's 5 rules and 5 permission checks.
No. Guard is local-first: interception, decisions, and evidence stay on your machine unless your workspace explicitly configures cloud features. This page is documentation only — it never executes a command or changes protection state.
Guard runs locally alongside the major AI coding agents and MCP-capable harnesses, so this coverage applies wherever Guard intercepts actions. See the supported harness guides at https://hol.org/guard/harnesses for per-tool approval behavior and limitations.
Guard has a free local tier that includes command interception, evidence, and the policy controls this listing documents. Team plans add shared policy, review queues, and audit surfaces. Current plans: https://hol.org/guard/pricing
Open an issue or pull request against the hol-guard repository, where the canonical catalog lives: https://github.com/hashgraph-online/hol-guard Listings are corrected through the same public review process that adds them.
Yes. Community extensions are merged through public pull requests with recorded provenance, and you can claim the publisher page for a contribution you maintain through the Publisher Studio.
The command mapping table is the audit trail for covered commands and catalog default floors. Pair it with the reviewed rule and permission counts, the per-tool state mapping, and the exact source tree linked from View exact source at the reviewed commit. Guard records enforcement decisions with evidence locally, so what ships matches what you reviewed.