Fact Check Skill Changelog

v3.1

Expanded references/educational-tips.md into a full media-literacy base: per-technique defenses with prebunking "vaccine" framing, a cognitive-biases reference, research citations, general tips (10-second rule, SIFT, lateral reading, image verification), and a native Bulgarian educational base for Bulgarian-language output.
Added references/bulgarian-context.md covering Bulgarian information-space narratives, recurring disinformation patterns, and local data points. Complements the source/search lists in bg-eu-sources.md.
Broadened the red-flag taxonomy in references/red-flags.md (more logical-fallacy and health-specific markers) while keeping the existing JSON severity scale and the red_flags[] schema unchanged.
SKILL.md: added an explicit Step 0 date-anchoring instruction, a satire/opinion guard in the non-negotiables, a corrections-and-updates (revisions) policy, and Bulgarian/English trigger phrases for better discovery.
Harmonized the example red-flag codes in the test fixtures with the expanded taxonomy. No change to the JSON schema, result validator, renderer, packaging, or test behavior.

Reworked the skill from one large SKILL.md into a concise progressive-disclosure structure with dedicated references/ files.
Added prompt-injection guardrails for untrusted user, screenshot, and fetched source content.
Added an evidence-led workflow with explicit evidence ledger fields, source independence tracking, and access limitations.
Added Bulgarian/EU source guidance for local fact-checking, EU policy, health, science, statistics, and geopolitical claims.
Added MFS calibration notes and clarified that MFS is a heuristic risk score, not a probability.
Added JSON result schema plus deterministic validation and HTML rendering scripts.
Added regression fixtures and unit tests for standard, comparison, and prebunking modes.
Added repository-level CI to validate skill metadata, run renderer tests, and build upload archives.
Removed committed generated .skill and .zip artifacts from source control; build them from scripts/package_skill.py instead.

Expanded lateral reading, source evaluation, MFS scoring, and educational sections in the previous monolithic skill file.

Added standard, comparison, prebunking, and quick-check modes.
Added source scoring, red-flag detection, origin tracing, confidence calibration, and share-safe summaries.

Initial fact-checking workflow with claim decomposition, source checking, red flags, verdicts, and educational tips.