> ## Documentation Index
> Fetch the complete documentation index at: https://docs.agentguardian.io/llms.txt
> Use this file to discover all available pages before exploring further.

# Roadmap

> Where AgentGuardian is going next — corpus growth, adapter coverage, report features, CI/CD integrations.

## How to read this

AgentGuardian is in active development. The roadmap below tracks the
public OSS package only; the Enterprise tier has its own roadmap
(out of scope for this site — see
[Open vs Enterprise](/concepts/open-vs-enterprise)).

The canonical source for the roadmap is the
[`docs/community/oss-roadmap.md`](https://github.com/glacien-technologies/agent-guardian/blob/main/docs/community/oss-roadmap.md)
file at the repo root, and the
[GitHub Projects board](https://github.com/orgs/glacien-technologies/projects)
for live progress.

<Info>
  Anything not in the OSS roadmap is **not** a public commitment. We don't
  ship features from marketing copy — we ship features from PRs.
</Info>

## Themes for the current cycle

### Probe corpus growth

The corpus is at **96 probes across 10 OWASP ASI 2026 categories**
(see `src/agent_guardian/probes/_meta/version.yaml`). The roadmap
covers:

* Expanding ASI04 (supply chain) with MCP-registry-specific probes as
  the MCP-server ecosystem matures.
* Expanding ASI10 (rogue agent / drift) with long-horizon probes that
  span multiple scan windows.
* New ASI09 (trust exploitation) probes for emerging output-channel
  attacks (artifact-rendering, agentic-document tampering).
* Continued alignment with new MITRE ATLAS releases.

### Target-adapter coverage

The current adapters live under `src/agent_guardian/adapters/framework/`
and `src/agent_guardian/transports/`. Roadmap items track:

* Deeper coverage of the OpenAI Agents SDK.
* Anthropic Computer-Use and tool-use coverage.
* Gemini-CLI agentic patterns.
* A2A v1.0 protocol coverage as it stabilises.
* Better default support for streaming responses across all adapters.

### Reports and exports

The current report engines live at `src/agent_guardian/reports/`:
`json`, `sarif`, `junit`, `markdown`, `pdf`, plus the signed `bundle`.
Roadmap items:

* Per-finding remediation rendering inline in the HTML preview.
* SARIF 2.1.1 stability across GitHub, GitLab, and Azure DevOps code-
  scanning views.
* Better JUnit per-suite grouping for CI failure tabs.

### CI/CD integrations

The goal is to make AgentGuardian a one-line install in the major CI
systems. Current native support: GitHub Actions, GitLab CI. Roadmap:

* A reusable GitHub composite action (no more raw `pip install` step).
* Pre-built GitLab CI template.
* CircleCI orb (community-driven).
* A `--diff-against` flag so PR scans compare to the main-branch
  baseline.

### Developer experience

* A `--profile fast-iteration` mode for tighter dev loops.
* Better terminal output for the scan progress view (richer per-agent
  status lines).
* A standalone `agent-guardian inspect <probe-id>` for probe debugging.

## How to influence the roadmap

The roadmap is **not** a closed list. The fastest paths to influence:

1. **Open a Discussion.** Use the
   [Roadmap category](https://github.com/glacien-technologies/agent-guardian/discussions/categories/roadmap)
   to propose a new direction or signal what you'd like to see
   prioritised.
2. **Open an issue with a concrete proposal.** A specific probe, a
   specific adapter, a specific report feature. We label
   `kind/roadmap-candidate` for those.
3. **Ship a PR.** A working PR is the most persuasive form of roadmap
   input we accept. See [Contributing](/community/contributing).

## What's explicitly *not* on the OSS roadmap

These are deliberate non-goals for the OSS package — they belong to the
Enterprise tier or to other products:

* **Runtime defensive controls.** AgentGuardian is a *testing*
  framework, not a *runtime gateway*. We do not ship policy
  enforcement, request blocking, or live traffic interception in OSS.
  See [Open vs Enterprise](/concepts/open-vs-enterprise).
* **Managed evidence storage.** Reports are written to disk locally.
  We do not host them.
* **Team / SSO / audit-log workflows.** No multi-user UI in OSS.
* **Telemetry.** AgentGuardian does not phone home — by design,
  not as an option to opt out of.

## Release cadence

We aim for a tagged PyPI release whenever the `CHANGELOG.md` has more
than a few user-visible entries — historically every 2-4 weeks. The
release process is documented in
[`MAINTAINERS.md`](https://github.com/glacien-technologies/agent-guardian/blob/main/MAINTAINERS.md).

## Next step

<CardGroup cols={2}>
  <Card title="Contributing" icon="users" href="/community/contributing">
    Six on-ramps for contributors — probe, adapter, eval, demo, docs, security.
  </Card>

  <Card title="Release notes" icon="clipboard-list" href="/community/release-notes">
    What shipped in the latest release and the one before it.
  </Card>
</CardGroup>