Prompts

Everything the routine loads at runtime. The two master prompts (daily-cti-brief.md, weekly-summary.md) drive every run; the supporting files (verification.md, brief-template.md, check-brief-fixes.md) are read by the prompts at runtime as policy, template, and remediation guide.

Prompts and runtime policies

Version history

Every substantive prompt edit ships with a CHANGELOG entry explaining why the editorial policy shifted between two committed briefs. Recent versions:

2.64 — 2026-06-20 (drop Wayback fallback repo-wide; remove the CVE Summary Table from briefs; RSS fetch-waste demotions)
2.63 — 2026-06-20 (every-run source-accessibility health check + recipe-aware probe; Ops dashboard floats only unsolved source problems)
2.62 — 2026-06-20 (source-metadata-drift correction made a first-class lifecycle step; per-run sources_changed telemetry surfaced on the Ops dashboard)
2.61 — 2026-06-20 (weekly aligned to the daily gold standard in structure + procedure, with a sharpened horizon lens; daily↔weekly division of labour made explicit)
2.60 — 2026-05-25 (weekly ISO-week anchor — most-recent-Sunday; fixes Sun→Mon-boundary duplicate weeklies)
2.59 — 2026-05-15 (work/<run-id>/ becomes version-controlled — every run commits the sub-agent artefact dir alongside the brief)
2.58 — 2026-05-15 (Tier 1 + 3 + 4 — source-quote binding, verifier convergence, workflow hardening)
2.57 — 2026-05-15 (Tier 2 mechanical pre-verifier checks: anchor-resolution, quantifier-evidence, tldr-body-drift, name-collision)
2.56 — 2026-05-15 (telemetry: fetch_failures[] becomes a strict "coverage gaps" log, bridge_uses[] tracks successful bridge calls separately, Ops dashboard sub-agent card rebuilt with labelled metrics)
2.55 — 2026-05-15 (bridge fetcher: generic feed <URL> subcommand for any RSS/Atom, 16-source verification with per-article drilldowns, RDF/Atom parser fix, 6 new sources added)

Full version history →