Changelog

Dated changes to the skills, installation metadata, and documentation. Verification limits accompany the relevant entries.

2026-10-28

Homepage and Sam’s Env, collection 0.12.0

Rebuilt the homepage with large Geist typography, a dark grid, restrained entrance motion, and accessible copy feedback. Every published skill card comes from the checked catalog, so new catalog entries appear automatically.
Renamed the canonical env-example package to sams-env, matching Sam’s Env’s current invocation. Existing /env-example/ documentation routes remain available; installation and invocation examples use the new skill name.
Added the environment-example spacing checker and formatter, including checks for explanatory blocks, annotation adjacency, and commented assignments. Clarified setting-specific dashboard links and app-secret versus provider-token guidance.
Phone Size was already published in collection 0.11.0; it requires no duplicate package or draft label. Commit, push, deployment, and live verification remain separate states.

2026-10-20

Skill Creator docs and a shared collection catalog

Split Skill Creator guidance into a first-use path, task examples, behavior and dependency reference, and verification and troubleshooting.
Publish the documentation research behind these changes, with primary Stripe, Cloudflare, and Google sources.
Drive complete skill listings and navigation from one catalog, checked against canonical skill directories and guide routes.
Add CI checks for missing skills, stale generated lists, route failures, docs tests, and the production build.
Keep skill instructions and plugin versions unchanged. Documentation checks do not establish new skill behavior or automatic discovery.

2026-10-16

Phone Size, collection 0.11.0

Add phone-size, a portable successor to phone-roll for concise task, decision, research and update content.
Replace fixed cards and link counts with content-specific briefs and destination-native disclosure.
Preserve evidence, uncertainty, scope, access and authority; distinguish content checks from actual rendered phone fit.
Include independent synthetic trials and their scoped results. Automatic runtime discovery, native Linear write behavior and physical iPhones remain untested.
Source publication and global installation do not imply documentation-site deployment.

2026-10-15

Tree Visual Markers, Collection 0.10.1

Require a meaningful emoji or visual marker on every tree row, including leaves and collapsed branches.
Keep category markers distinct from work status; explicit no-emoji requests use compact text markers.
Record the October 14 synthetic forward check and document repository-backed symlink installations for local development.
Source publication does not imply a documentation deployment.

2026-10-12

Sam’s Skill Creator, collection 0.10.0

Added sams-skill-creator, a single authoring workflow that consults relevant methods at each decision.
Locked Anthropic Skill Creator, Writing for Agents, Automate Me, and Zach Prompting, with a separate Codex packaging snapshot and Poteto runtime support.
Added frozen restores, read-only freshness checks, integrity tests, bundled synthetic cases, and honest baseline comparisons.
Kept review-only tasks read-only and made history mining conditional on preference-based work.
Recorded bounded authoring/review validation; automatic discovery and downstream use of generated skills remain untested.

2026-10-10

Tree and Status, Collection 0.9.0

Added independently installable sams-tree and sams-status skills.
Kept familiar groupings, three displayed levels, collapsed branches, and -> paths as the shared presentation defaults.
Added phase-relative status, Needs Info, brief parent summaries, likely-stale markers, and source links outside the tree.
Captured the confirmed design, glossary, synthetic evaluation requests, and verification evidence.

2026-10-09

Env Example and Collection 0.8.0

Added env-example with pragmatic setup guidance and standard static @env-spec comments.
Added a dependency-free Node.js checker, 11 behavioral tests, and a concise synthetic example.
Restricted existing env input to committed, Git-tracked .env.example blobs; actual env files remain excluded.
Added source-backed research, usage docs, and verification evidence. The checker intentionally supports a bounded static subset.

2026-10-07

Sam’s Gen Writing and collection 0.7.0

Added sams-gen-writing for drafting, reviewing, and planning AI-assisted web content for Google Search.
Grounded it in Search Central’s generative AI, helpful content, and spam policy pages (checked October 7, 2026) and in speaker-attributed Search Off the Record remarks from 2021 to July 2026.
Used Zach Prompting to keep one rule per behavior, explicit evidence tiers, no fabricated authors or experience, and four completion statuses.
Added five synthetic smoke scenarios. No behavioral pass rate is measured yet; see verification.

2026-10-02

Andrej Explain mechanical checks, collection 0.6.1

Added local Node text and SRT checkers with JSON reports and meaningful exit codes.
Enforce explicit text contracts, semicolon checks, caption structure, ordering, overlap, and optional duration limits.
Keep approximate STE length findings advisory and distinguish all script results from full compliance or semantic verification.
Added boundary and CLI regression tests and wired the checks into the skill’s verification step.

Andrej Explain and collection 0.6.0

Added andrej-explain for source-grounded explanations of ideas, code, research, and agent work.
Merged both research threads, recovered the original post link with its indirect verification limit, and corrected the STE chart’s TEST entry.
Used Zach Prompting to preserve the explanation contract while making media, controlled-language rules, and learning checks conditional.
Added an install-to-first-result guide, examples, behavior reference, and verification record informed by Stripe’s documentation patterns.
Added retained forward-use fixtures and package checks. See verification for observed results and limits; website deployment is separate.

2026-10-01

Trunk-based contributions

Merged the migration workflow and installation documentation into main.
Made direct-to-main publication the default for ordinary authorized changes; explicit review requests, contributor access limits, and repository rules still take precedence.
Kept validation, authorization, and no-force-push requirements. No branch protection, security, or permission settings were changed.
Updated current installation instructions to use the published default branch rather than an unmerged PR preview.

Backup-first GPT migration and collection 0.5.0 candidate

Added private original-detail capture, checksummed file inventory, and explicit missing-file states before conversion.
Documented the standard Agent Skills folder, local/repository delivery, and account-dependent ChatGPT Web plugin installation.
Added a hand-test checklist; live owned-GPT parity and Web installation remain untested.
Removed private workspace provenance from maintained documentation, aligned bundle versions and migration metadata, and stabilized the evaluation link.
Hardened the recorded-evidence checker for optimized Python and pinned the current requirements retrospectively without claiming historical pre-run proof.

2026-09-26

Custom GPT migration evaluations

Added six synthetic migration fixtures with explicit expected requirements, retained conversion artifacts, and generated-skill response checks.
Separated conversion inspection, task execution, destination delivery, and automatic discovery in the migration verification guidance.
Recovered and ran the original docs-spec gate through its documented setup.
Recorded current plugin, project-installation, skill, and documentation checks with live-integration and native-runtime limits. No deployment was performed.

2026-09-22

Zach Prompting Sol and Luna guidance

Added verified GPT-6 Sol and Luna API details and a conditional prompt migration checklist.
Shortened skill activation metadata while preserving the newer Astra, session-authority, and delegation rules.
Clarified approved preparation before outstanding action gates and source checks for exact model and runtime settings.
Added representative comparison criteria; Sol and Luna behavior improvements remain awaiting evals. GitHub publication and website deployment are separate states.

2026-09-18

Sales Research and Collection 0.4.0

Packaged Sales Research as a self-contained skill with practitioner routing, primary-source links, and communication guidance.
Removed private-workspace and companion-skill requirements from the portable version.
Added installation instructions, seven prompt templates, a source guide, and a tester checklist. Existing synthetic scenarios remain unscored criteria.
Updated collection discovery and plugin metadata. GitHub publication and website deployment remain separate from local validation.

2026-09-17

Custom GPT migration

Added migrate-cgpt for portable conversion to local agents, personal skills, and GitHub repositories.
Preserve required outputs and authorization rules; track missing knowledge and Action integrations explicitly.
Added conditional delivery, component mapping, verification, and dated source references.
Added a guide and verification record. Documentation deployment remains a separate step.

2026-09-14

Install command preferences

Simplified the dropdown to tool names, hover states, and the selected checkmark; removed group headings and descriptions.
Replaced the native selector with a custom ChatGPT-style menu, selected-item checkmarks, keyboard navigation, and viewport-aware positioning.
Reduced the choices to npx, bunx, pnx, mise, and Yarn dlx. Migrate saved aliases without losing the runner preference.
Added a shared runner picker for Skills CLI command blocks, with npx, bunx, pnpx, pnpm dlx, Yarn dlx, npm exec, mise, and pnx options.
Remember the latest selection locally and synchronize it across command blocks, pages, and open tabs.
Copy the selected command with its environment variables and multiline flags intact. Preserve the original npx examples for clients without JavaScript and the agent-readable docs.
Disable code-font ligatures and preserve inline whitespace to make spaces before flags visible.

Sam UX Audit and Collection 0.3.0

Added independently installable sam-ux-audit for websites and GitHub repositories.
Bundled the core audit method and pinned five upstream review dependencies with frozen installation and integrity checks.
Added public smoke results for W3C before/after pages, GOV.UK, TodoMVC and Express. Findings distinguish observed behavior from source risks.
Kept the stable marketplace install key and sam namespace. The GitHub package includes documentation source; website deployment is a separate action.

2026-09-05

Documentation editing

Applied p-unslop to maintained documentation, skill instructions, references, and the memo template. Replaced repeated introductions, abstract wording, and dense sentences with direct explanations.
Preserved commands, schemas, citations, requirements, and recorded evaluation evidence.

Zach Prompting Astra update

Added conditional GPT-6 Astra guidance and dated source attribution.
Clarified authorization carryover, instruction-conflict reporting, continuation after follow-ups, and proportional verification.
Preserved the existing review, rewrite, debug, and migration modes and standalone install name.
Behavioral improvement remains awaiting comparative evals; structural checks do not establish model performance.

2026-07-17

0.2.1

Relocated the repository-wide skill and instruction system audit and its remaining roadmap to Harness Atlas, while keeping the HTMA behavior contract in this collection.
Strengthened HTMA Measure so missing value-of-information inputs remain unknown or explicitly qualitative instead of receiving numerical defaults.
Added the VOI score state, nullable expected value, and missing-input contract to the feature and architecture documentation.
Recorded the remaining audit sequence on the project roadmap.

2026-07-15

HTMA Measure problem examples

Added a source-backed guide to documented estimation pains, including budget-before-quote decisions, point-estimate risk, hidden work, target-price ambiguity, small samples, research stopping rules, and asymmetric error costs.
Added copyable prompts that use explicit bracketed fields rather than fabricated measurements or realistic-looking placeholder results.
Added a human-facing README beside the canonical HTMA Measure skill and aligned the root repository guide and Starlight navigation.

HTMA Measure evaluation suite

Ported the original generic HTMA scoring utilities and scrubbed aggregate H1-H12 history beside the skill.
Added a transparent ten-case H12-derived regression packet, deterministic scorer, isolated Claude CLI runner, and recorded native-agent baseline.
Used Zach Prompting to reconcile estimate-versus-block transitions, then compared the unchanged skill and two isolated candidates on the same regression suite.
Added an explicitly synthetic six-case H13 acceptance suite for missing-threshold, current-lookup, ambiguous-target, and unavailable-private-actual behavior.
Updated HTMA Measure so a missing threshold blocks only the action comparison when a numeric range remains responsible, while missing identifiers, current sources, and private actuals still block numeric output.
Preserved the private boundary for source ledgers, raw historical outputs, mappings, human score sheets, and active sealed holdouts.
Reclassified the sanitized H12 packet as a visible regression set; its results are not fresh holdout evidence.

Sam’s Skills collection

Reframed the repository and documentation site as the collection of skills Sam uses and shares.
Added collection-level About and Installation pages, and moved collection navigation ahead of skill-specific sections.
Changed the bundled Claude Code namespace from htma-measure to sam while retaining the stable marketplace install key.
Verified public commit 8405b24 through a fresh HTTPS marketplace install, strict manifest validation, component inventory, and recursive skill-content comparison.

Zach Prompting

Added the standalone zach-prompting skill for reviewing, rewriting, debugging, and migrating prompts and agent instruction artifacts.
Added conditional GPT-5.6 and Claude Fable 5 guidance sourced from the official OpenAI and Anthropic prompting guides.
Added four fresh-context forward uses covering rewrite, review-only, model migration, and a real repository instruction audit; recorded their scope and limitations on the verification page.
Added the public Zach Prompting guide, examples, source attribution, and Skills CLI installation command.
Verified the published GitHub commit with clean remote Skills CLI and Claude marketplace installs.

Documentation collection

Reframed the root README, docs landing page, navigation, and repository reference around a multi-skill collection while preserving the existing HTMA Measure documentation.

0.2.0

Added Zach Prompting to the root skills/ collection.
Reframed the existing backward-compatible htma-measure marketplace plugin as the two-skill ThatGuySam Skills bundle.
Preserved the existing Codex and Claude installation ID while documenting the bundled namespace and the one-skill Skills CLI path.

2026-07-11

Documentation site

Added the public Starlight documentation application at apps/docs.
Added installation, usage, prompt, command, troubleshooting, feature, architecture, and reference pages.
Added public llms.txt, llms-full.txt, and llms-small.txt outputs.
Deployed the site at https://skills.samcarlton.com.

0.1.1

Changed Claude’s plugin source to the same marketplace repository.
Removed the second GitHub SSH clone requirement.
Documented the complete two-command Codex installation.

0.1.0

Published the standalone htma-measure skill.
Added Codex, Claude Code, and portable plugin manifests.
Added the calibrated memo template, method map, output rubric, and local paid-quote adjustment.
Sanitized personal and workspace-specific markers for public use.