Radar / Ideas / SkillRoster: review agent skills once…

SkillRoster: review agent skills once and verify the exact version installed

weekly debate · 2026-W44moderateJEV confidence 0.56JEV monetization –2026-10-30

OutcomeReduce installs of unapproved skill versions to zero through the pilot’s managed install path, measured against request and install logs.

The problem

SkillSpector scans skills before installation, while shared agent workspaces need a way to reuse review decisions. Neither a scan alone nor an approval of an earlier version establishes that the installed package was reviewed.

The idea

Build a tool-agnostic catalog that hashes submitted packages, saves SkillSpector findings and human decisions, and emits an approval manifest. A local installer checks the manifest against the package hash before copying it into an administrator-selected OpenBot workspace.

Why now

SkillSpector accepts skill packages for pre-install review, and OpenBot offers persistent team workspaces where approved packages could be reused.

What it combines

SkillSpector evaluates a package; OpenBot provides a recurring team-use setting. A hash-bound decision lets workspace administrators reuse a review without silently approving changed code.

How to build it

Accept packages. Upload a folder or zip and calculate its content hash. (1 day)
Scan and review. Save SkillSpector findings and a named administrator’s decision against the hash. (1 day)
Issue approval manifests. Record approved hashes and reviewer decisions in a verifiable local manifest. (1 day)
Pilot managed installs. Check the package hash before installation into an administrator-selected workspace; log requests and disclose bypasses. (1 day)

Who benefits

Workspace administrators: Can identify precisely which package version was reviewed.
Agent users: Can reuse reviewed packages through a managed path.

How it makes money

Model: Per-team subscription
Pricing: $149 per team per month for the pilot
Wedge: One team administrator managing shared skills
Channels: Direct outreach to engineering teams · Open-source project communities

Distribution

Engineering or security teams pay. Start with an administrator willing to use the managed installer for one shared workspace; reach teams through direct outreach and the projects’ communities.

Why it wins

A scanner produces findings; SkillRoster ties those findings and a named reviewer’s decision to the exact bytes distributed through a managed install path.

Risks

OpenBot may not expose a documented install gate, and users can bypass the installer. The MVP uses a separate administrator-run installer and reports managed-path coverage, not organization-wide enforcement.

Strongest objection left standing

Without control over every installation route, the catalog cannot prevent determined users from installing unreviewed skills.

Build it with

NVIDIA SkillSpector: security scanner for AI agent skills, trending againProduces pre-install scan findings for submitted packages.
OpenBot: local-first desktop workspace where a team of AI agents works togetherProvides the pilot’s shared agent-workspace use case without assuming a native install gate.

Repo to start from

skillroster - hash-bound skill approvals and managed installation

Evidence

Debated by openai/gpt-6-sol, anthropic/claude-sonnet-5.5, qwen/qwen3.8-max-0902. Judged by JEV.

Get the week's best AI launches, plus 3 ideas worth building

One email every Saturday. Ranked by traction, not hype. Free.