agenda-intelligence-md 0.6.1

A drop-in markdown cognition layer for AI agents that need to analyze public agenda

Agenda Intelligence MD

A machine-readable Markdown protocol and CLI toolkit that turns news scanning into decision-ready agenda analysis.

Bottom line: agents using this protocol move from “monitor developments” to “watch these 3 indicators; if X happens, decision Y becomes urgent.”

---

Why this exists

Most LLM news workflows summarize what happened. Agenda Intelligence asks the harder operational questions: what changed, whether it is noise or signal, who gains leverage, what evidence supports the claim, what remains uncertain, and what to watch next.

It is useful when an agent needs to reason about public agenda rather than produce a generic brief: policy, sanctions, regulation, geopolitics, energy, elections, markets, and strategic risk.

Who it is for

• Policy analysts & think‑tanks – structured monitoring of regulatory and geopolitical signals.
• Sanctions & compliance teams – evidence‑backed briefs that satisfy audit requirements.
• Market‑risk analysts – concise impact snapshots for faster investment decisions.
• Founders & operators – external chaos turned into clear strategic inputs.
• AI agents – any LLM‑based system that needs to analyze public agenda without drifting into generic summaries.

---

Before / after

Before: "The regulator announced new AI guidance. Companies should monitor developments."

After:
- Signal classification: compliance-relevant development
- What changed: policy discussion moved toward implementation detail
- Main uncertainty: whether guidance becomes the practical enforcement baseline
- Watch next: regulator guidance, first enforcement action, compliance deadline
- Evidence status: source-backed / reasoning-only / unsupported claims separated

See examples/before-after/ and examples/source-backed/ for full examples.

---

Quickstart

# 1. Install
pip install agenda-intelligence-md

# 2. Start with a source category
agenda-intelligence start technology-ai

# 3. Validate and score a source-aware brief
agenda-intelligence validate-brief examples/agenda-brief.json
agenda-intelligence score examples/agenda-brief.json --evidence examples/source/evidence-pack.json

# Optional: print local MCP client config
agenda-intelligence mcp-config

Expected scoring output:

score: 90/100
note: Heuristic structural/evidence-discipline score; does not verify factual truthfulness.
evidence_support: ... claims supported: 1/1 supported ...

Proof points in this repo:

• schemas/ validates agenda briefs, evidence packs, memory cards, lenses, and signal classification.
• SOURCE_POLICY.md defines the evidence discipline and overclaiming guardrails.
• examples/source-backed/ shows evidence-mode-aware briefs.
• docs/evaluation.md describes the evaluation rubric, judge prompt, and regression checks.
• CI runs validation, scoring, schema checks, and public example smoke tests.

Agent instruction block

Copy this into an agent system prompt, project instructions file, or tool-specific rule block:

When the task involves news, policy, regulation, sanctions, geopolitics, trade, energy, elections, conflicts, markets, or strategic risk, use Agenda-Intelligence.md.

Do not summarize by default. Classify the signal, identify what changed, separate fact from assessment, name assumptions and unknowns, explain who gains or loses leverage, and end with watch-next indicators.

Use SOURCE_POLICY.md and source-requirements/<category>.json before making high-stakes claims. Never imply live verification unless live verification was performed. If evidence is missing, label the claim as unsupported or reasoning-only.

See docs/integrations/agent-instruction-block.md for a fuller copy-paste block.

Other install modes

# From GitHub Release
pip install https://github.com/vassiliylakhonin/agenda-intelligence-md/releases/download/v0.6.1/agenda_intelligence_md-0.6.1-py3-none-any.whl

# Editable install from source
git clone https://github.com/vassiliylakhonin/agenda-intelligence-md
cd agenda-intelligence-md
pip install -e .

---

CLI happy path (aha‑moment in 5‑10 min)

The start command is the primary onboarding entry‑point. It prints a trimmed source plan, a brief template, and next commands.

# 1. Onboard with a source category
agenda-intelligence start technology-ai

# Output: trimmed source plan + brief template + next commands

# 2. Validate a bundled example brief
agenda-intelligence validate-brief examples/agenda-brief.json

# 3. Score a JSON brief (structural quality check)
agenda-intelligence score examples/agenda-brief.json

# 4. Add an evidence pack for claim-level evidence discipline
agenda-intelligence score examples/agenda-brief.json --evidence examples/source/evidence-pack.json

# 5. Or score a before/after markdown example
agenda-intelligence score examples/before-after/eu-ai-act.md

Scorer status: the score command now scores JSON briefs with a heuristic 0-100 structural rubric, can include an evidence pack for claim-level support feedback, and still supports before/after markdown examples. It does not verify factual truthfulness.

Demo output (what you see after start):

=== Trimmed source plan ===
{
  "must_check": ["tech‑release", "policy‑update", "market‑data"],
  "watch_indicators": ["regulation draft", "enforcement action"]
}

=== Brief template (fill in) ===
{
  "bottom_line": "<summary>",
  "signal_classification": "<signal>",
  "what_changed": "<what changed>",
  "main_uncertainty": "<main uncertainty>",
  "watch_next": ["<indicator 1>", "<indicator 2>"]
}

---

Protocol · Source Policy · Schemas

Layer	Purpose	Status
Markdown protocol (Agenda-Intelligence.md)	Core reasoning workflow (signal classification, watch‑next, etc.)	✅ Stable
Source Acquisition Layer	Tells the agent which source types to check before making claims (sanctions, regulation, elections, conflict, etc.)	✅ Stable
JSON schemas	Validate briefs, evidence packs, memory cards	✅ Stable
AnalysisBank	Memory layer that stores reusable reasoning patterns from good/bad outputs	✅ Stable
Regional & Sector lenses	Central Asia & Caspian, Middle East, EU; sanctions, export controls	✅ Stable

---

Stable today vs Experimental

✅ Stable today

• Markdown protocol (Agenda-Intelligence.md) – core reasoning workflow.
• JSON schemas – validation for briefs, evidence packs, memory cards.
• CLI validation – validate-brief, validate-evidence, validate-manifest.
• Source plans – source-plan, list-source-packs, source-types.
• Guided start – start command prints trimmed plan + brief template.
• Brief scoring – score examples/agenda-brief.json returns a heuristic 0-100 structural quality score; add --evidence for claim-level evidence feedback.
• Evaluation toolkit – evals/rubric.md, evals/llm_judge_prompt.txt, evals/human_checklist.md, evals/cases/*.json.
• MCP stdio server – agenda-intelligence-mcp exposes validate_brief, validate_evidence, get_protocol, list_lenses, get_lens, source_plan.

🧪 Experimental / Planned

• MCP HTTP/WebSocket transport – stdio is available; HTTP/WebSocket transport remains on the roadmap (docs/integrations/mcp.md).
• Fetch command – stub in CLI, full evidence‑pack retrieval not implemented.
• Truthfulness evaluation – not implemented; current scoring checks structure and protocol signals.
• Generate‑brief – not yet exposed; use start + manual template fill.

Note: The project is young. Stable parts are ready for production use; experimental bits are usable for testing but may change.

---

Examples

• Source‑backed briefs (with evidence mode & source plans):
  examples/source-backed/eu-ai-act.md, sanctions-routing.md, red-sea-shipping.md
• Classic examples: examples/hormuz_strait_brief.md, eu-brief.md, central-asia-caspian-brief.md
• Before/after: examples/before-after/ – shows the delta when the protocol is applied.

---

Documentation

Resource	Link
Quickstart	docs/quickstart.md
End‑to‑end tutorial	docs/tutorial.md
Evaluation assets	evals/ – rubric, LLM judge prompt, human checklist, sample cases
Use‑cases	docs/use-cases/ – policy monitoring, sanctions compliance, market risk, founder context
Integrations	docs/integrations/ – Claude Code, OpenAI Codex, Cursor, MCP
Release maintenance	docs/release.md
Roadmap	ROADMAP.md
Changelog	CHANGELOG.md

---

Repository structure

agenda-intelligence-md/
├─ src/agenda_intelligence/   # Python package
├─ schemas/                   # JSON schemas
├─ examples/                  # sample briefs, evidence packs, source‑backed examples
├─ analysis-bank/             # memory cards (failures & successes)
├─ evals/                     # evaluation rubric, LLM judge, human checklist, cases
├─ docs/                      # guides, tutorials, use‑cases, integrations
├─ skills/agenda-intelligence/ # OpenClaw skill wrapper
└─ tests/                     # pytest suite

---

Roadmap (high‑level)

• v0.6 – MCP stdio server exposing current Python tool functions.
• v0.7 – Promote score_output to rubric-based quality scoring and wire it into MCP.
• v0.8 – Automated CI quality gate using the evaluation toolkit.
• v1.0 – Stable API, broader adoption‑channel support, production‑grade MCP.

---

Contributing

Pull requests are welcome! Please:

1. Open an issue to discuss changes.
2. Create a feature branch (feat/..., fix/...).
3. Run pytest and ensure all tests pass.
4. Update docs if behavior changes.

---

License

MIT – free for any use.

---

Why this exists

Most agent‑written news analysis is a polished recap that doesn’t change any decision. This project gives agents a stricter workflow so their output actually helps someone decide, hedge, or act.