Weekly Intelligence Digest

weekly-intelligence-digest

The flagship scheduled agent — the one the user explicitly asked for in the original ideation. Every Monday morning, it surfaces what changed in the past 7 days across the sources that matter for SDLC modernization at a regulated organization, framed in plain language with the compliance-relevant lens.

Agent config (consumed by scheduled-agent-runner)

agent_name: weekly-intelligence-digest
cadence: weekly
schedule_hint: "Monday 07:00 local"
source_filter:
  topic_tags:
    - github
    - copilot
    - ghas
    - frontier-model
    - ai-coding-tools
    - regulator
    - ai-governance
    - supply-chain
    - compliance-framework
  min_tier: 2          # tier 1 + tier 2 — skip unverified tier-3
  max_items_per_source: 5
verify_loadbearing: true       # Top-3 items get verify-claim (3-vote refute)
curate_findings: true          # Stage promotable claims to _inbox/ for memory-curator
digest_template_overrides:
  why_you_care_extra: |
    Frame every "Why You Care" line through the lens of: SDLC modernization for a
    large engineering organization, GitHub admin operations (Copilot/CodeQL/Dependabot/GHAS),
    or US financial regulation (SR 11-7 model risk, FFIEC, OCC, NYDFS 500, SOX ITGC).
    If a finding doesn't tie to one of these, drop it from the digest — it belongs in
    research/ or a different surface, not the weekly intelligence brief.

Intent

A Monday-morning brief the user reads in 5 minutes:

1. What changed last week that they need to know about — Copilot release, Anthropic model launch, FFIEC guidance, OpenSSF advisory affecting the org's stack.
2. Why they care — the compliance-relevant implication, not the marketing copy.
3. Sources with credibility tier badges so claims are auditable months later.

This is the canonical example of Use Case 1 — Scheduled weekly research.

TL;DR ranking rules

Pick the top 3 items for the TL;DR by signal:

1. Regulator action that mandates or strongly suggests a control change.
2. Vendor product change that changes the answer to a Copilot-FAQ-style question the user fields regularly (data handling, audit log, content exclusion, IP indemnity).
3. Frontier-model release with implications for which models Copilot/Cursor/etc. will route to.
4. Supply-chain incident or advisory affecting the org's declared tech stack.

If fewer than 3 high-signal items, leave TL;DR with fewer bullets — don't pad with filler.

"Why You Care" examples

Good (concrete, compliance-relevant):

Copilot Enterprise audit log S3 export — Regulated buyers can now ingest into SIEM via direct S3 + KMS, replacing scheduled manual CSV exports. Reduces audit-evidence gathering effort and brings Copilot activity into the standard pipeline alongside the rest of the org's logs.

Bad (vague, vendor-tone):

Copilot got better at security. Worth checking out.

The "Why You Care" framing connects vendor-speak to the org's actual operational reality.

Output surface

vault/digests/weekly/YYYY-MM-DD-weekly-intelligence-digest.md where YYYY-MM-DD is the run date (period_end).

Composes with

All Phase-1 foundation skills via the runner. This skill itself is mostly config + framing.

Acceptance test (for step 7 done-criteria)

One end-to-end run. Confirm:

• Digest file lands at the path above.
• All 5 sections present (TL;DR / What Changed / Why You Care / Detailed Findings / Sources).
• Tier badges render correctly.
• Failures from unverified sources appear in Sources (don't silently omit).
• .state/weekly-intelligence-digest/seen.jsonl populated.
• At least one source surfaces real items.