Launch an agent with scoped permissions
A reviewed tool inventory, tested abuse cases and a controlled release decision.
Preview
What this package looks like once your agent has it.
# Download the complete evaluated bundle: curl -fL https://superagentskill.com/api/public/plugins/solution-secure-agent-launch.zip -o solution-secure-agent-launch.zip # Version: 0.1.31 # Evaluated runtime: typesafe/jev-1.14.0/security-source-state-v1@node-24.22.0-arm64/compiled-playbook-gateway-v29-structural-record-identities # Unpack and follow solution-secure-agent-launch/skills/solution-secure-agent-launch/SKILL.md. # Keep its runtime profile and explicit source-upload consent. # Prompt copying or MCP access alone is not this evaluated execution.
Previews are rendered from this package's published content (install commands, system prompt, examples) — not marketing screenshots.
Description
A reviewed tool inventory, tested abuse cases and a controlled release decision.
Package source (inspection)
Inspect the source here. The complete evaluated execution requires the bundled runner and runtime profile.
slug: solution-secure-agent-launch
name: Launch an agent with scoped permissions
type: playbook
version: 0.1.31
description: A reviewed tool inventory, tested abuse cases and a controlled release decision.
trigger: When the user requests launch an agent with scoped permissions and provides authorized
source inputs.
inputs:
source_1:
type: string
description: Tool manifests and data classifications
source_2:
type: string
description: Permission policy and authorized test scope
source_3:
type: string
description: Synthetic fixtures and release criteria
outputs:
deliverable:
type: object
description: Every tool has an owner and minimum scope; Abuse tests distinguish observed results
from plans; Release owner accepts remaining risks
steps:
- id: inventory
action: tool:security.inventory
with:
requires: []
- id: fixtures
action: tool:security.fixtures
with:
requires:
- inventory
- id: test
action: tool:security.test
with:
requires:
- fixtures
- id: release
action: tool:security.release
with:
requires:
- test
tags:
- Agent security
- business-solution
- candidate
license: CC-BY-SA-4.0
authors:
- Super Agent Skill
x_solution_id: secure-agent-launch
x_context_budget: 4
x_package_dependencies:
- mcp-server-builder
- prompt-injection-tester
- iam-least-privilege
- security-review
Stats
Adoption, reception and performance for this package.
No published performance metrics yet.
Install instructions
Run the complete package in its evaluated environment.
Use the evaluated bundle
Version 0.1.31 was evaluated with its packaged runner and the environment below. Connecting an MCP client or copying the prompt does not reproduce that execution.
typesafe/jev-1.14.0/security-source-state-v1@node-24.22.0-arm64/compiled-playbook-gateway-v29-structural-record-identitiesDownload evaluated ZIP v0.1.31Verify archive integrity with the signature sidecar and public key.
- Unpack the ZIP and follow
solution-secure-agent-launch/skills/solution-secure-agent-launch/SKILL.md. - Keep the included runtime profile, dependencies and exact model version. Other models and environments remain unqualified.
- Configure
TYPESAFE_API_KEYprivately in your local environment. The runner sends authorized input fields to TypeSafe only with explicit source-upload consent. Use the bundled instructions to configure the pinned Node host and, where required, the XML reader.
The evaluation covers source-grounded preparation. Financial actions, external sending, real security tests and release approval require separate authorized evidence.
Changelog
Recorded versions, newest first. Qualification remains version-specific.
Versions
| Version | Status | Released | Notes | |
|---|---|---|---|---|
| 0.1.31 | stable | 7 days ago | ||
| 0.1.0 | Outside current qualification | 9 days ago |
Compatibility
Qualification is restricted to this version’s complete bundle and the exact evaluated environment shown below. It does not transfer to bare prompts, other models or external actions.
Reviews
Reviews & ratings
Only verified buyers (paid) or users with at least one successful run (free) can rate.