Skip to content

CHANGELOG.md

Latest commit

 

History

History
116 lines (79 loc) · 5.72 KB

CHANGELOG.md

File metadata and controls

116 lines (79 loc) · 5.72 KB
Raw

Changelog

All notable changes to MindFluence are documented here. The format follows Keep a Changelog and the project uses Semantic Versioning.

[2.5.1] - 2026-11-07

Changed

  • install.py copies LICENSE into the installed skill and includes it in --check; scripts/ is shipped with the skill.

[2.5.0] - 2026-11-07

Added

  • scripts/anti_patterns.py: deterministic checks for AP-1...AP-13, --tier, --json, [FAIL]/[REVIEW].
  • tests/test_anti_patterns.py: 13 adversarial samples, a compliant sample, and same-paragraph checks for AP-2/AP-5.
  • CI runs the anti-pattern tests; install.py ships scripts/.

Changed

  • Post-Generation Verification now points to the mechanical lint.

[2.4.0] - 2026-11-07

Changed

  • Moved the runtime skill to skills/mindfluence/ to match the suite layout.
  • Moved reference files to skills/mindfluence/references/.
  • Added install.py for Claude Code, OpenCode, Codex, and the generic .agents convention.
  • Added Claude and Cursor plugin manifests, .editorconfig, and an OpenAI agent manifest.
  • Frontmatter normalized: quoted description, added metadata.repository, environment-based compatibility, metadata.author.
  • Section headings normalized to Title Case; H1 aligned with the suite.
  • Version normalized to three-part semver (2.3.0).
  • CI now validates frontmatter, version, relative links, and the installer plan.
  • Reference files are read by relative path from the skill folder instead of GitHub URLs.
  • Long dashes replaced with hyphens across all documentation.
  • Section canon: added When Not to Use, renamed the output section to Output Contract.
  • Canonical placeholder vocabulary documented: [N], [DATA NEEDED: ...], [VERIFY: ...], [Source, Year], [illustrative].

[2.3.0] - 2026-10-30

Added

  • PROVENANCE / TRUTH GATE as mandatory post-generation check #7: every number, name, quote, testimonial, study, journal, and year must be traceable to user input, a verifiable source, or an explicit placeholder.
  • Data provenance policy: if data is missing, use placeholders ([N], [Customer name], [Source, Year], [measurable result]) and a [DATA NEEDED] note, or ask the user in Deep Mode. Never invent specifics to satisfy a formatting check.
  • Two compliant narrative modes: Mode A - real-data NPSA (full name + location/role + timeline + verbatim quote + visceral detail); Mode B - labeled [illustrative] first-person founder narrative or generalized vignette.
  • SECURITY.md with vulnerability reporting guidance, scope, and supported versions.
  • CHANGELOG.md (this file).
  • Placeholder guidance in AP-1, AP-4, AP-13 and in the verification checks.
  • [DATA NEEDED] blocks in STANDARD, EXTENDED, and Rewrite output formats.

Changed

  • Completed all 38 Bias Selection Router rows so each row covers Filter + Optimizer + Social categories.
  • Post-generation verification expanded from 6 to 7 checks; output templates and all documentation now reference the 7-check gate.
  • Narrative requirements now accept Mode A or Mode B; the verification check no longer forces a fabricated named person.
  • AP-5 clarifies the cold-outreach exception: value and ask must be separated, and the value must be genuinely standalone.
  • All inlined and external examples are now explicitly marked as illustrative; scenario-level AP-1/AP-4 guidance accepts placeholders when data is unavailable.
  • README and README.ru updated for v2.3: 38-row router, 7 checks, provenance, two narrative modes, changelog/security links.

Fixed

  • ANTI-PATTERNS (12) heading corrected to 13; anti-patterns.md now includes AP-13 and an audit checklist covering all 13 anti-patterns plus provenance.
  • RskReversal typo corrected to RiskReversal across the scenario table.
  • Bias Blind Spot example no longer violates AP-10 (Insulting Confirmation).
  • Unclear "Gilmore Girls" unexpected-detail example replaced with a generic niche-reference example.
  • "4 Hofstede dimensions" wording corrected to Hofstede + Hall's high/low-context dimension.
  • README claims about the router size (50 → 38), verification count (6 → 7), and anti-pattern count (12 → 13) corrected.
  • README tree now includes README.ru.md, CHANGELOG.md, and SECURITY.md.
  • OpenCode installation wording corrected: discovery is path-based, not frontmatter-based.

Security

  • Documented the security scope of a prompt-only skill, including prompt injection, fabricated citations, unsafe marketing use, and accidental secret exposure.
  • Clarified that _ship/ is local tooling and must remain out of version control.

[2.2] - 2026-07-12

Added

  • Self-contained single-file architecture with inlined Bias Selection Router, Scenario Quick-Reference, Cultural Quick-Reference, 13 anti-patterns, 14 power combinations, and 7 tone styles.
  • Named Person Story Arc (NPSA) narrative requirement with fallback patterns.
  • Rewrite Without Metrics variant with 7 rewrite intents.
  • 3-tier output format (COMPACT / STANDARD / EXTENDED) and 6-point post-generation verification.

Changed

  • Replaced the 10-step bias-selection procedure with a single router lookup.
  • Added narrative minimums per tone and keyword density caps.

[2.1] - 2026-07-12

Added

  • Production hardening: refusal policy, error handling, typo sweep, cultural depth.

[2.0.2] - 2026-07-12

Fixed

  • Patch release after v2.0.1.

[2.0.1] - 2026-07-12

Fixed

  • Patch release after v2.0.

[2.0] - 2026-07-12

Added

  • Major rewrite of the skill structure, modes, and scenario coverage.

[1.3] - 2026-06-25

Added

  • Production hardening: refusal policy, error handling, typo sweep, cultural depth.
  • MIT license and repository cleanup completed in the preceding release cycle.