Changelog — Next Move Theory canon & skills

What changed in the Next Move Theory canon + skills. Newest release is at the top.

Each entry's Summary line is what the installed skills show you when a newer version is available. To update to the latest at any time, run from your project root:

curl -fsSL https://raw.githubusercontent.com/zamesin/Next-Move-Theory-Canon-and-Skills/main/install.sh | bash

or use the /nmt-update skill (Codex: $nmt-update). Windows (PowerShell): irm https://raw.githubusercontent.com/zamesin/Next-Move-Theory-Canon-and-Skills/main/install.ps1 | iex.

Versioning

One version covers the whole bundle (canon + skills), tracking Next Move Theory as SemVer MAJOR.MINOR.PATCH:

MAJOR — a major version of the Next Move Theory methodology.
MINOR — a significant methodology update (new or reworked theses).
PATCH — small skill updates and methodology patches.

The current version is the top entry below. The installer records it in .nmt-version; at the end of a run each skill checks whether a newer version is published and tells you what changed since yours. The check sends only the skill name and the installed version, and it is opt-out — see Updates & telemetry in the README. Updates themselves always download from the public GitHub repo. (The README also shows the methodology maturity badges — Advanced JTBD v3.4 · stable and Next Move Theory v0.6 · in active development.)

0.6.18 — Transparent updates, GitHub-based installs, and an anti-hallucination pack

Summary: Update checks are now transparent and one-line opt-out, updates download from GitHub, /nmt-upgrade became /nmt-update, the rules block injected into your CLAUDE.md shrank to five lines, and the producer skills got a hard anti-hallucination pack — frequency counts, claim statuses, segmentation guards, and honesty about thin input. Released: 2026-09-13

Transparent, opt-out update checks. The check runs at the end of a run, announces itself in one line the first time, and sends only the skill name and the installed version. Turn it off with update-check: off in .nmt-config, or DO_NOT_TRACK=1, or NMT_NO_UPDATE_CHECK=1. The new Updates & telemetry section of the README states exactly what is and isn't collected.
Updates come from GitHub. The installer, the one-liners, and the version lookup all read the public repo (zamesin/Next-Move-Theory-Canon-and-Skills) directly — no site round-trip in the update path.
nmt-upgrade is now nmt-update (/nmt-update, Codex $nmt-update). The installer removes the old skill folder and installs the new one, so you won't end up with both.
A much smaller footprint in your rules file. The block injected between the Next-Move-Theory-Rules markers in CLAUDE.md / AGENTS.md is five lines: where the canon is, start with /nmt-chat, where outputs go, where the full guide is. Your own content outside the markers is untouched, as before.
Anti-hallucination pack in the producer skills. Every aggregated claim now carries a frequency count and its source references, and a single-source claim can no longer become a segment-level conclusion — it goes to a Single signals list. Every claim is tagged as backed by a quote, derived from the data, or the model's own hypothesis; numbers a respondent said out loud are labelled as that person's opinion, not market fact; ideas the model invented live only in Hypotheses to validate. Segmentation runs through an explicit guard list: purchase channel is not a segment, industry is not a segment when the Core Jobs and success criteria coincide, and every segment is defined through Core Jobs + success criteria. Interview analysis ships a source-to-segment appendix so every "16 deals in this segment" total can be re-checked line by line.
Honesty scaled to your input. With thin input the report says so in the header, gives ranges instead of point numbers, and lists the top three inputs that would most improve the answer.
Market and language are asked up front. Examples, channels, price anchors, competitors, and tone all come from your market — no silent US or Russian default.
Repo context is read only with permission. The skills list the context files they found in your folder and ask before opening them, with a plain notice that the content is processed locally and sent nowhere.
Plain tooltips in HTML output. Every abbreviation and methodology term is explained on first use, in words that assume no methodology knowledge.

0.6.17 — Plain-language skills + a clearer first run

Summary: Skills now lead in plain words (methodology terms in parentheses), point you to /nmt-chat when you're unsure where to start, ask "quick vs thorough" up front, and keep answers shorter. No methodology change. Released: 2026-06-24

Plain-first wording across every skill: an everyday explanation leads, the methodology term follows in parentheses — or is used directly for common words like segment, Aha moment, and problem. All glosses were re-verified against the canon.
Every skill now opens with a "New here? Start here" pointer to /nmt-chat, so you always know which skill to run first.
The producer skills ask how deep to go up front — a few key questions for a fast pass, or the full interview for the highest-confidence result.
Shorter by default: the one-page answer leads; the deeper work is opt-in, not a wall.
nmt-craft-value-proposition and nmt-product-requirements no longer ask you to write a Job in formal notation — describe it in plain words and the skill structures it for you.
The post-install message now points you straight to /nmt-chat and maps every skill.

0.6.16 — Faster launch-time update check

Summary: The skills' update check now runs inline instead of reading a separate file, so skills start a touch faster. No change to what the check does. Released: 2026-06-22

Inlined the update check into each skill's body — removed the separate VERSION-CHECK.md read on launch; behaviour unchanged (best-effort, ≤2s, never blocks).
Trimmed the Codex skill descriptions to sit well under the platform's description-length limit.

0.6.15 — Auto-update checks + one-command upgrade

Summary: Skills now check on launch whether a newer version is available and show what changed since yours. Added the /nmt-upgrade one-command updater (Codex: $nmt-upgrade). Released: 2026-06-22

New nmt-upgrade skill — re-runs the official installer to pull the latest canon + skills from the public repo, in place and idempotently.
Every skill now runs a lightweight, best-effort version check on launch (≤2s, silent on failure, never blocks): if your installed version is behind, it tells you the gap and the main changes per version, then how to update.
The installer now records the installed version in .nmt-version so the check can compare it against this changelog.

0.6.14 — nmt- prefix, Claude/Codex split, Windows installer

Summary: All skills moved to the nmt- prefix with separate Claude (/nmt-…) and Codex ($nmt-…) variants, and a Windows PowerShell installer was added. Released: 2026-06-22

Skills republished as two agent variants under Skills/{claude,codex}: seven nmt-prefixed skills (nmt-chat, nmt-diagnose, nmt-market-research, nmt-craft-value-proposition, nmt-product-requirements, nmt-craft-go-to-market, nmt-analyze-interviews).
Claude uses /nmt-… and installs to .claude/skills; Codex uses $nmt-… and installs to .agents/skills.
Added install.ps1 (Windows PowerShell) and nmt-chat zero-state onboarding (paste what you have → next move).

0.6.13 — One-command installer + the diagnose skill

Summary: Added the one-command curl … | bash installer and a new nmt-diagnose front-door skill, alongside shared producer and readability contracts. Released: 2026-06-20

One-command installer (install.sh) that clones the repo and installs the canon, skills, and rules block into your project root, idempotently.
New nmt-diagnose skill — a conversational front door for live products that surfaces risks, growth points, and risky assumptions, then routes to the right producer skill.
Producer + readability contracts shared across the producer skills.