Skip to content

Agenda Intel MD

v0.9.1 Feature

This release adds 3 notable features for engineering teams evaluating rollout.

Published 13d MCP Developer Tools
✓ No known CVEs patched
Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

a2a a2a-protocol agenda-intelligence agent-infrastructure agentic-workflow ai-agents
+14 more
cloudflare-workers deal-risk-gate decision-grade geopolitical-intelligence geopolitical-risk json-schema mcp mcp-server model-context-protocol policy-analysis regulatory-compliance risk-analysis sanctions strategic-intelligence

ReleasePort's take

Light signal
editorial:auto 13d

v0.9.1 fixes EU routing gap observed in GTTA evaluation and adds optional audience_detail field for context preservation. No breaking changes; unit test suite expanded to 121 tests.

Why it matters: EU routing fix requires validation if you route Europe terms. Adopt audience_detail to preserve caller intent at boundaries. No breaking changes; safe next regular upgrade.

Summary

AI summary

Updates audience_detail` optional freeform field, Tooling, and exact across a mixed release.

Changes in this release

Feature Medium

`route_modules` now routes EU / Europe terms to `eu` regional specialist.

`route_modules` now routes EU / Europe terms to `eu` regional specialist.

Source: granite4.1:8b-q6_K@2026-05-21

Confidence: high

 —
Feature Medium

`MODULE_PATHS["eu"]` resolves to existing eu.md reference module.

`MODULE_PATHS["eu"]` resolves to existing eu.md reference module.

Source: granite4.1:8b-q6_K@2026-05-21

Confidence: high

 —
Feature Medium

Added optional `audience_detail` string field to request schema.

Added optional `audience_detail` string field to request schema.

Source: granite4.1:8b-q6_K@2026-05-21

Confidence: high

 —
Feature Medium

`_format_request_context` renders `audience_detail` when present.

`_format_request_context` renders `audience_detail` when present.

Source: granite4.1:8b-q6_K@2026-05-21

Confidence: high

 —
Dependency Medium

No new dependencies; only additive changes in v0.9.1.

No new dependencies; only additive changes in v0.9.1.

Source: granite4.1:8b-q6_K@2026-05-21

Confidence: low

 —
Performance Medium

Test count increased from 111 to 121 unit tests.

Test count increased from 111 to 121 unit tests.

Source: granite4.1:8b-q6_K@2026-05-21

Confidence: low

 —
Bugfix Medium

Resolved silent caller-intent loss at audience boundary.

Resolved silent caller-intent loss at audience boundary.

Source: granite4.1:8b-q6_K@2026-05-21

Confidence: high

 —
Bugfix Medium

Fixed EU routing gap observed in GTTA evaluation.

Fixed EU routing gap observed in GTTA evaluation.

Source: granite4.1:8b-q6_K@2026-05-21

Confidence: low

 —
Bugfix Medium

Resolved EU routing gap identified in GTTA evaluation (observation 3).

Resolved EU routing gap identified in GTTA evaluation (observation 3).

Source: granite4.1:30b@2026-05-21-audit

Confidence: low

 —
Refactor Medium

`Makefile` now includes `PYTHON ?= python3` override for test/ci commands.

`Makefile` now includes `PYTHON ?= python3` override for test/ci commands.

Source: granite4.1:8b-q6_K@2026-05-21

Confidence: low

 —
Refactor Low

`Makefile` now defines `PYTHON ?= python3` to allow overriding Python interpreter for tests and CI.

`Makefile` now defines `PYTHON ?= python3` to allow overriding Python interpreter for tests and CI.

Source: granite4.1:30b@2026-05-21-audit

Confidence: low

 —
Full changelog

What changed in v0.9.1

Patch release covering post-v0.9.0 work merged on main. Additive only — no new schemas, no new MCP tools, no live retrieval. Existing v0.9.0 requests remain valid.

EU geography and term routing

  • route_modules in src/agenda_intelligence/product.py now routes geography "EU" / "Europe" (exact) and question text containing long-form EU terms (european union, european commission, eu ai act, gdpr, cbam, cjeu, nis2, brussels, ecb, schrems, etc.) to the eu regional specialist. The bare two-letter "eu" is matched only as an exact geography token to avoid false-positive substring hits inside words like exposure or queue.
  • MODULE_PATHS["eu"] resolves to the existing skills/agenda-intelligence/references/regional/eu.md.
  • Closes the gap surfaced by evals/agent-eval/gtta-global-policy.md observation 3: the GTTA SKILL listed EU as a loadable lens but route_modules had no EU branch.
  • Seven new unit tests covering exact-geography, phrase-in-question, substring false-positive regression guard, global-only-routes-GTTA, EU + sanctions composition, and module content loadability.

audience_detail optional freeform field (ADR 0009)

  • schemas/agenda-request.schema.json (and packaged copy) gains an optional audience_detail string (minLength: 1). The existing enum-bound audience field stays as the prompt-routable signal; audience_detail carries the caller's original framing (e.g. "Refinery sanctions compliance and trade-finance leadership").
  • _format_request_context renders audience_detail into the verified request-context block alongside audience when present.
  • Closes the second observation from evals/agent-eval/gtta-global-policy.md about silent caller-intent loss at the audience boundary.

Three completed agent-eval cases

All three v0.9 scaffold cases are now complete and committed under evals/agent-eval/:

| Case | Skill under test | Score A | Score B | Delta |
|---|---|---|---|---|
| gtta-global-policy.md | global-think-tank-analyst | 3.5/8 | 8/8 | +4.5 |
| ca-caspian-sanctions.md | central-asia-caspian-hybrid-intelligence-skill | 3.5/8 | 8/8 | +4.5 |
| gulf-me-hormuz-shipping.md | gulf-middle-east-hybrid-intelligence-skill | 4.0/8 | 8/8 | +4.0 |

Same model (Sonnet 4.6 adaptive) in both conditions per the methodology. Delta consistent in the +4.0 to +4.5 range across three independent topics. B scores 8/8 in all three. First production validation that the product shell can route canonical regional-skill rules (e.g. Gulf+ME's Iran-state / IRGC-affiliated / Iran-private commercial actor distinction) through the assembled system prompt.

Tooling

  • Makefile gains a PYTHON ?= python3 override variable so make test and make ci work on machines where the system python3 lacks dev deps. GitHub Actions behavior unchanged.

Test count

111 → 121 (+10 new unit tests across EU routing and audience_detail).

Install

pip install agenda-intelligence-md==0.9.1

Full changelog: CHANGELOG.md

View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track Agenda Intel MD

Get notified when new releases ship.

Sign up free

About Agenda Intel MD

All releases →

Related context

Earlier breaking changes

  • v0.8.0 MCP tool count increased from 11 to 16, adding five new tools.

Beta — feedback welcome: [email protected]