Skip to content
Agent Threat Rules (ATR)中文

Implementer Report

Implementer report

A specification counts once more than one engine implements it the same way. ATR does not have that evidence yet. This report lists pull requests that other projects merged, all opened by the ATR maintainer. They show where ATR rules ended up. They do not show that anyone built an engine from the spec alone.

Working Draft·spec 1.0.1 (dated 19 October 2026)·npm 4.3.1·page generated 20 November 2026·canonical /spec·editor Adam Lin

No row in this report claims a conformance level. See how the levels are defined

Current implementers

Each row is a pull request merged into another organization's public repository. The ATR maintainer opened every one of them, and that project's maintainers reviewed and merged it. No implementation here has run the conformance suite, so none of these is a conformance claim. A merged pull request shows the integration happened. It does not show it is still in use. Every row can be checked on GitHub.

Microsoft Agent Governance Toolkit

What was merged
Example directory: an adapter that runs ATR rules through the ACS annotator
Opened by
ATR maintainer
Merged (UTC)
2026-06-15 @imran-siddique
PR #3025

Cisco AI Defense (skill-scanner)

What was merged
ATR rule pack. On 2026-11-02 a maintainer wrote that they will likely remove ATR because the false-positive rate is too high. A report-only pack (#289) was closed unmerged on 2026-11-05.
Opened by
ATR maintainer
Merged (UTC)
2026-04-22 @vineethsai7
PR #99

MISP (CIRCL), galaxy

What was merged
ATR galaxy and cluster
Opened by
ATR maintainer
Merged (UTC)
2026-05-10 @adulau
galaxy #1207

MISP (CIRCL), taxonomies

What was merged
agent-threat-rules taxonomy
Opened by
ATR maintainer
Merged (UTC)
2026-05-10 @adulau
taxonomies #323

OWASP Agent Security Regression Harness

What was merged
4 regression scenarios derived from ATR rules. The scenario files do not name ATR.
Opened by
ATR maintainer
Merged (UTC)
2026-05-11 @mertsatilmaz
PR #74

Gen Digital Sage

What was merged
Agent-layer threat patterns in threats/agent-layer.yaml
Opened by
ATR maintainer
Merged (UTC)
2026-05-11 @vaclavbelak
PR #33

This table is kept by hand. The full list of adopters is ADOPTERS.md. Maintainers do not pre-approve adopters. A PR that fits the schema and links to public evidence anyone can verify is enough to merge. ADOPTERS.md

Standards-body submissions (open, not adoptions)

ATR offered two AI RMF profile examples to NIST's usnistgov/oscal-content repo as PR #338. NIST added its own AI RMF catalog to the develop branch in July, and the PR was reworked against it in August. It is still open, waiting for NIST review. This is not a NIST adoption or endorsement, so it is not in the implementers table above. #338

How to self-certify

ATR uses self-certification. Conformance is shown by running the tests, not by saying so. Implementers run the conformance test suite locally, then open a pull request against ADOPTERS.md with their integration metadata. There is no review board and no paid certification. No one has self-certified this way yet. The minimal procedure has three steps.

  1. Run the conformance test suite locally. The test suite is under conformance/v1.0/. Each fixture is a directory named after the rule, with the input in input.yaml and the expected result in expect.json. An implementation passes when every fixture comes out as expected. The reference engine does not pass all of them yet; the suite's README lists the ones that fail.
  2. Open a pull request against ADOPTERS.md. Submit against ADOPTERS.md on the main branch. Maintainers usually validate the schema and merge within seven days.
  3. Include full integration metadata in the PR. At minimum: the organization name, the spec version your integration targets, the integration date (ISO 8601), a public evidence link anyone can check (a PR, a commit or a public documentation URL), and the conformance level you claim (L1 Engine, L2 Publisher or L3 Sub-range Authority).

The maintainers reserve the right to re-run the test suite against any published self-declaration at any time, to check the declared result. The TSC is being formed and takes this over once seated.

Conformance levels at a glance Informative

This section is informative. The normative text for each level is at /conformance.

L1 engine conformance

The L1 engine is what makes the format portable. It MUST parse every rule that validates against spec/atr-schema.yaml. It MUST evaluate detection.conditions with the semantics in §3.5, and it MUST honor the scan_target and status semantics. That way a rule gets the same verdict no matter who wrote the engine. An L1 engine MAY refuse rules outside the scan_target it declares.

L2 publisher conformance

An L2 publisher puts out rules in its own vendor-prefixed range, such as ACME-YYYY-NNNNN, and follows every ATR semantic. It MUST honor the deprecation policy in §3.7 and SHOULD include test_cases with every rule it publishes.

L3 sub-range authority

An L3 sub-range authority is a national or organizational body that issues rules under a sovereign prefix, such as ATR-TW-2026-NNNNN. Authority is granted through the procedure in /charter §5. Once the TSC is seated, it grants it. Until then the lead maintainer (BDFL) does.

Read the full normative text: the test-suite structure, the self-certification procedure, and how conformance gets verified.


Editor: Adam Lin <[email protected]> — DOI 10.5281/zenodo.19178002 — MIT License — ISO 8601 2026-10-19