Agent Skills Platform: Governed Workflow Packaging for AI Agents
Build tested agent skills and govern their lifecycle through a user-defined marketplace: evidence, discovery, updates, rollback, quarantine, and 17-platform distribution.
At a glance
- What is it?
- Agent Skills Platform is an MIT-licensed Python project that turns recurring expert workflows into tested, installable skill packages for AI agents, with a marketplace layer that handles versioned releases, rollback, and team distribution. It targets organisations that need governance over what their agents are authorised to do.
- Who is it for?
- Agent Skills Platform is worth evaluating for teams that have recurring expert workflows and need to share them across colleagues in a controlled way, particularly when those workflows touch APIs, databases, or structured data that requires a pinned contract. Solo developers with a single workflow have no need for the marketplace layer and can use the private path instead.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 16 days ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on October 1, 2026, and from our analysis. They are not legal advice.
Editorial analysis
The Problem Agent Skills Platform Addresses
When a subject matter expert completes a recurring task with an AI agent, the prompt they developed, the evidence behind their decisions, and the quality checks that validate the output are typically lost. The next person who runs the same task starts from scratch or copies a prompt without any of the reasoning that shaped it.
Agent Skills Platform treats a workflow as a first-class artifact: a tested, installable package that carries the business question, the evidence behind the work, functional scripts where needed, evals, security checks, and a representative run record. The README describes the output as not just a prompt but an installable skill package with a correction path for real-world learning.
The platform targets three distinct roles. A subject matter expert supplies examples and approves the result. A creator uses an AI agent to build and verify the skill. A marketplace operator governs publication, versioning, rollback, and quarantine. Teammates who need the skill install an approved version without touching the creation or governance machinery.
How a Skill Package Is Assembled
Creating a skill starts with the workflow owner describing the task in natural language and attaching examples of the work, which might be spreadsheets, reports, emails, screenshots, or scripts. The README gives a concrete example of a monthly revenue-variance review:
/agent-skills-platform
Turn my monthly revenue-variance review into a reusable internal skill.
I attached past reports and the source spreadsheets. The decision is whether to
escalate a material variance. It must not modify source data.The creator agent asks for the business decisions only the expert can authorise, then builds and tests the skill and shows a representative result. When the workflow connects to an API, database, MCP server, codebase, or structured file, a process called Semantic Recon runs automatically before implementation to create a pinned data contract.
The README distinguishes between what runs at creation time and what runs at execution time: classified run evidence goes into a raw/ directory, recurring findings become evidence-linked draft patterns in a wiki/ directory, and only a separately validated change may update the executable skill. The README explicitly states this is a governed maintenance record, not autonomous self-modification.
Installation and First Skill
The platform installs as a Claude Code plugin through the marketplace or manually by cloning the repository. The README shows both paths.
Plugin installation:
claude plugin marketplace add FrancyJGLisboa/agent-skills-platform
claude plugin install agent-skills-platformManual installation by cloning:
git clone https://github.com/FrancyJGLisboa/agent-skills-platform.git
cd agent-skills-platformFor an installation that skips the Semantic Recon step, the README shows:
./install.sh --without-semantic-reconThis flag is described as appropriate only for a deliberately local, source-free installation where no external data contract is needed.
For teammates who are not technical, the README describes a desktop app that requires no terminal, Python, or Git. Users download the app, paste a team library link, and click Install. Skills land in GitHub Copilot or another supported tool. The README also notes that a governed team marketplace must be created before skills can be shared; it requires a GitHub or GitLab repository and takes four commands to set up.
The Verification Record and Evidence Layer
The repository includes a live, read-only weather briefing example that illustrates the expected output of the platform. The README points to a verification record document and a skill package directory in references/examples/live-weather-briefing-skill/ as the canonical demonstration.
The platform distinguishes between knowledge, capability, and execution. The README states that a skill is not itself a RAG system, an MCP server, or an agent runtime. RAG supplies knowledge. MCP supplies capabilities. The harness supplies execution. A skill organises those components into a governed path toward a verified outcome. This distinction matters for teams evaluating whether the platform replaces or sits on top of their existing infrastructure.
The evidence management design is a specific architectural choice: maintenance evidence is kept separate from the concise instructions the agent executes at runtime, so the running prompt does not grow with every correction or failed experiment. The README frames this as a way to let skills learn from real-world use without accumulating prompt context that degrades performance.
Where the Platform Has Limits
Agent Skills Platform has no GitHub releases. The repository description mentions 17-platform distribution, but the README does not enumerate all 17 platforms, so the full list requires reading the installation documentation separately.
The platform requires an AI agent environment to create skills; the README says to open the AI agent you already use and paste the prompt. This means the quality of skill creation depends on the underlying agent's capability. The README does not specify which agent environments are supported beyond Claude Code and Codex for the reliability certification feature.
The governance structure adds overhead that is not appropriate for solo developers or one-off tasks. The README explicitly states: use the team path only when teammates will install or reuse the skill itself. Sending a report or queue to a colleague does not require a marketplace. The added complexity of ownership records, approval workflows, and versioned releases only pays off when multiple people are consuming the same skill repeatedly.
Agent Skills Platform vs. Direct Prompt Engineering
The most direct alternative to Agent Skills Platform is maintaining a shared library of raw prompts in a Git repository, which is what most teams currently do. The difference in approach is testability and governance. A raw prompt file carries no evals, no representative run record, no security checks, and no rollback mechanism. Anyone can edit it. There is no record of which version was in use when an output was produced.
Agent Skills Platform adds all of those controls at the cost of installation overhead and a dependency on an AI agent to build skills. It also requires a marketplace operator to govern team distribution, which is an organisational role that does not exist in a simple prompt library.
For teams where prompt quality and accountability matter, and where the same workflow runs repeatedly across multiple people, the platform's structure addresses real risks that a flat prompt file does not. For individual developers running one-off experiments, a plain prompt file is the faster path and the platform's overhead is not justified.
Editorial conclusion
Agent Skills Platform is worth evaluating for teams that have recurring expert workflows and need to share them across colleagues in a controlled way, particularly when those workflows touch APIs, databases, or structured data that requires a pinned contract. Solo developers with a single workflow have no need for the marketplace layer and can use the private path instead. The no-releases repository means you are working from the main branch; review the GATES.md and CHANGELOG.md files before deploying to a production team environment.
Frequently asked questions
What is an agent platform?
Agent Skills Platform defines an agent platform as an environment where tested, installable skills govern how agents move from a recognised situation to a verified outcome, with separate layers for knowledge (RAG), capabilities (MCP), and execution (the harness). The platform itself does not replace those layers but organises them into a governed workflow package.
Can you give me some examples of agent skills?
The README gives a monthly revenue-variance review as a worked skill example: the skill captures the escalation decision logic, the source spreadsheets as evidence, and a representative run record. The repository also includes a live weather briefing skill in references/examples/live-weather-briefing-skill/ with a full verification record.
Does Agent Skills Platform require a marketplace to use?
No. The README describes a private path for creating, verifying, and installing a skill just for yourself, with no marketplace setup required. The team marketplace path is only needed when teammates will install or reuse the skill itself.
Official sources
Add this badge to your README
If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.
[](https://hysenlabs.com/projects/francyjglisboa-agent-skills-platform)