Report #001 — do skill verdicts hold across a model release?

Release drift report

v1.2 (2026-07-30) — this edition supersedes v1.1. An in-text grounding policy was codified and applied across all 10 suites; imported rubric criteria in git-workflow-and-versioning and commit-work were amended and those two suites re-run. See Amendments.

Model pair: claude-sonnet-5 (released 2026-06-30) vs claude-sonnet-4-6 (released 2026-02-17)
Judge: claude-haiku-4-5, 5 samples/case (surface-controlled) · Surface: claude-cli · Skills: 10

9 of 10 skills moved beyond noise on this model pair.

2 regressed · 3 improved · 4 mixed · 1 within noise — across 9 case regression(s) and 8 case improvement(s). A verdict is claimed only when BOTH (1) two with_skill bands (mean ± stddev over judge samples) do not overlap AND (2) the mean moves at least the 0.05 effect floor; bands that overlap, or separate by less than the floor, are reported as within noise, never as drift.

Per-skill summary

skill	source	with_skill (old → new)	baseline (old → new)	verdict
`code-review-and-quality`	addyosmani · MIT	0.869 ± 0.025 → 0.854 ± 0.031	0.824 → 0.836	WITHIN NOISE
`git-workflow-and-versioning`	addyosmani · MIT	0.868 ± 0.025 → 0.834 ± 0.104	0.788 → 0.859	MIXED
`documentation-and-adrs`	addyosmani · MIT	0.825 ± 0.100 → 0.706 ± 0.297	0.809 → 0.720	MIXED
`commit-work`	Softaworks · MIT	0.825 ± 0.082 → 0.849 ± 0.028	0.818 → 0.766	IMPROVED (1)
`writing-clearly-and-concisely`	Softaworks · MIT	0.864 ± 0.016 → 0.841 ± 0.064	0.839 → 0.813	REGRESSED (1)
`crafting-effective-readmes`	Softaworks · MIT	0.870 ± 0.015 → 0.804 ± 0.192	0.849 → 0.811	REGRESSED (2)
`naming-analyzer`	Softaworks · MIT	0.817 ± 0.077 → 0.839 ± 0.026	0.703 → 0.757	MIXED
`requesting-code-review`	Jesse Vincent (obra) · MIT	0.668 ± 0.249 → 0.719 ± 0.226	0.540 → 0.617	IMPROVED (1)
`writing-plans`	Jesse Vincent (obra) · MIT	0.811 ± 0.089 → 0.785 ± 0.145	0.576 → 0.675	MIXED
`skill-creator`	Anthropic, PBC · Apache-2.0	0.842 ± 0.095 → 0.875 ± 0.012	0.828 → 0.774	IMPROVED (1)

`code-review-and-quality` — Code Review and Quality WITHIN NOISE

A five-axis code-review framework and severity-labeling discipline for assessing changes before merge.

Source: https://github.com/addyosmani/agent-skills @ 7829ffd90d · License: MIT · Author: addyosmani
Receipts: old · new · drift

	`claude-sonnet-4-6` (old)	`claude-sonnet-5` (new)
with_skill (mean ± sd)	0.869 ± 0.025	0.854 ± 0.031
baseline (mean ± sd)	0.824 ± 0.081	0.836 ± 0.051

case	old (mean ± sd)	new (mean ± sd)	Δ	verdict
`severity-labeled-findings`	0.908 ± 0.042	0.794 ± 0.091	-0.114	within noise
`security-axis-sql-injection`	0.882 ± 0.008	0.858 ± 0.008	-0.024	within noise (below floor)
`commit-message-imperative-body`	0.862 ± 0.013	0.856 ± 0.009	-0.006	within noise
`reject-clean-it-up-later`	0.866 ± 0.015	0.868 ± 0.008	+0.002	within noise
`oversized-change-split`	0.856 ± 0.015	0.860 ± 0.010	+0.004	within noise
`propose-structural-remedy`	0.878 ± 0.019	0.896 ± 0.022	+0.018	within noise
`approve-when-improves-health`	0.828 ± 0.028	0.848 ± 0.026	+0.020	within noise

`git-workflow-and-versioning` — Git Workflow and Versioning MIXED

Trunk-based git discipline, atomic commits, conventional commit types, and semantic-versioning / changelog conventions.

Source: https://github.com/addyosmani/agent-skills @ 7829ffd90d · License: MIT · Author: addyosmani
Receipts: old · new · drift

	`claude-sonnet-4-6` (old)	`claude-sonnet-5` (new)
with_skill (mean ± sd)	0.868 ± 0.025	0.834 ± 0.104
baseline (mean ± sd)	0.788 ± 0.194	0.859 ± 0.024

case	old (mean ± sd)	new (mean ± sd)	Δ	verdict
`commit-message-conventional-type`	0.856 ± 0.013	0.600 ± 0.000	-0.256	regressed
`release-cut-version-tag-changelog`	0.910 ± 0.017	0.888 ± 0.011	-0.022	within noise
`changelog-curated-by-impact`	0.888 ± 0.029	0.876 ± 0.011	-0.012	within noise
`split-into-atomic-commits`	0.868 ± 0.008	0.858 ± 0.011	-0.010	within noise
`semver-clean-bump`	0.860 ± 0.012	0.856 ± 0.009	-0.004	within noise
`trunk-based-short-lived-branches`	0.864 ± 0.023	0.868 ± 0.013	+0.004	within noise
`semver-hidden-breaking-change`	0.832 ± 0.020	0.894 ± 0.022	+0.062	improved

`documentation-and-adrs` — Documentation and ADRs MIXED

When and how to write Architecture Decision Records, inline comments, and API docs; document the why.

Source: https://github.com/addyosmani/agent-skills @ 7829ffd90d · License: MIT · Author: addyosmani
Receipts: old · new · drift

	`claude-sonnet-4-6` (old)	`claude-sonnet-5` (new)
with_skill (mean ± sd)	0.825 ± 0.100	0.706 ± 0.297
baseline (mean ± sd)	0.809 ± 0.100	0.720 ± 0.293

case	old (mean ± sd)	new (mean ± sd)	Δ	verdict
`match-existing-adr-convention`	0.858 ± 0.016	0.076 ± 0.043	-0.782	regressed
`surface-conflicting-adr-conventions`	0.850 ± 0.025	0.582 ± 0.149	-0.268	regressed
`supersede-not-delete-old-adr`	0.886 ± 0.013	0.876 ± 0.005	-0.010	within noise
`comment-intent-not-implementation`	0.860 ± 0.014	0.864 ± 0.017	+0.004	within noise
`document-why-not-what-rewrite`	0.860 ± 0.010	0.868 ± 0.013	+0.008	within noise
`adr-for-costly-to-reverse-decision`	0.864 ± 0.018	0.876 ± 0.005	+0.012	within noise
`document-public-api-function`	0.600 ± 0.000	0.800 ± 0.000	+0.200	improved

`commit-work` — Commit Work IMPROVED (1)

Producing high-quality git commits: staging discipline, logical splitting, Conventional Commits messages.

Source: https://github.com/softaworks/agent-toolkit @ 3027f20f31 · License: MIT · Author: Softaworks
Receipts: old · new · drift

	`claude-sonnet-4-6` (old)	`claude-sonnet-5` (new)
with_skill (mean ± sd)	0.825 ± 0.082	0.849 ± 0.028
baseline (mean ± sd)	0.818 ± 0.063	0.766 ± 0.222

case	old (mean ± sd)	new (mean ± sd)	Δ	verdict
`full-workflow-multi-concern-diff`	0.906 ± 0.034	0.876 ± 0.011	-0.030	within noise
`patch-stage-mixed-single-file`	0.872 ± 0.008	0.860 ± 0.010	-0.012	within noise
`review-cached-catch-secret-and-debug`	0.852 ± 0.054	0.844 ± 0.092	-0.008	within noise
`split-dependency-bump-vs-behavior`	0.872 ± 0.004	0.866 ± 0.015	-0.006	within noise
`split-feature-vs-refactor`	0.852 ± 0.011	0.854 ± 0.011	+0.002	within noise
`conventional-commit-single-change`	0.738 ± 0.069	0.790 ± 0.041	+0.052	within noise
`two-sentence-describability-test`	0.682 ± 0.054	0.850 ± 0.023	+0.168	improved

`writing-clearly-and-concisely` — Writing Clearly and Concisely REGRESSED (1)

Strunk-style prose rules for docs, commit / error messages, reports, and UI text.

Source: https://github.com/softaworks/agent-toolkit @ 3027f20f31 · License: MIT · Author: Softaworks
Receipts: old · new · drift

	`claude-sonnet-4-6` (old)	`claude-sonnet-5` (new)
with_skill (mean ± sd)	0.864 ± 0.016	0.841 ± 0.064
baseline (mean ± sd)	0.839 ± 0.044	0.813 ± 0.100

case	old (mean ± sd)	new (mean ± sd)	Δ	verdict
`concrete-language-incident-summary`	0.862 ± 0.022	0.710 ± 0.073	-0.152	regressed
`emphatic-word-at-end`	0.848 ± 0.023	0.806 ± 0.013	-0.042	within noise (below floor)
`tighten-puffy-release-note`	0.874 ± 0.032	0.862 ± 0.022	-0.012	within noise
`positive-form-status-sentences`	0.886 ± 0.038	0.884 ± 0.015	-0.002	within noise
`active-voice-logging-passage`	0.852 ± 0.004	0.862 ± 0.011	+0.010	within noise
`omit-needless-words-cache-note`	0.878 ± 0.033	0.894 ± 0.039	+0.016	within noise
`keep-related-words-together-modifiers`	0.846 ± 0.011	0.872 ± 0.023	+0.026	within noise

`crafting-effective-readmes` — Crafting Effective READMEs REGRESSED (2)

Audience-first README structure with mandatory sections and project-type templates.

Source: https://github.com/softaworks/agent-toolkit @ 3027f20f31 · License: MIT · Author: Softaworks
Receipts: old · new · drift

	`claude-sonnet-4-6` (old)	`claude-sonnet-5` (new)
with_skill (mean ± sd)	0.870 ± 0.015	0.804 ± 0.192
baseline (mean ± sd)	0.849 ± 0.067	0.811 ± 0.144

case	old (mean ± sd)	new (mean ± sd)	Δ	verdict
`lead-with-one-sentence-problem`	0.874 ± 0.005	0.374 ± 0.136	-0.500	regressed
`categorize-task-before-writing`	0.868 ± 0.008	0.818 ± 0.034	-0.050	regressed
`audience-config-folder-future-you`	0.876 ± 0.015	0.870 ± 0.012	-0.006	within noise
`review-validate-against-project-files`	0.888 ± 0.029	0.890 ± 0.024	+0.002	within noise
`three-mandatory-sections-cli-tool`	0.872 ± 0.008	0.876 ± 0.013	+0.004	within noise
`audience-internal-service-runbook`	0.872 ± 0.004	0.894 ± 0.028	+0.022	within noise
`oss-project-type-sections`	0.840 ± 0.090	0.908 ± 0.041	+0.068	within noise

`naming-analyzer` — Naming Analyzer MIXED

Variable / function / class naming conventions and how to suggest clearer names.

Source: https://github.com/softaworks/agent-toolkit @ 3027f20f31 · License: MIT · Author: Softaworks
Receipts: old · new · drift

	`claude-sonnet-4-6` (old)	`claude-sonnet-5` (new)
with_skill (mean ± sd)	0.817 ± 0.077	0.839 ± 0.026
baseline (mean ± sd)	0.703 ± 0.214	0.757 ± 0.141

case	old (mean ± sd)	new (mean ± sd)	Δ	verdict
`go-acronym-casing`	0.868 ± 0.016	0.814 ± 0.022	-0.054	regressed
`constants-include-units-js`	0.866 ± 0.011	0.814 ± 0.022	-0.052	regressed
`vague-names-js`	0.854 ± 0.045	0.816 ± 0.042	-0.038	within noise
`language-casing-python`	0.886 ± 0.024	0.882 ± 0.013	-0.004	within noise
`misleading-name-mutation-js`	0.828 ± 0.008	0.842 ± 0.004	+0.014	within noise (below floor)
`boolean-prefixes-js`	0.720 ± 0.115	0.860 ± 0.010	+0.140	improved
`abbreviations-wellknown-js`	0.696 ± 0.099	0.846 ± 0.027	+0.150	improved

`requesting-code-review` — Requesting Code Review IMPROVED (1)

How to prepare and package a change for review: context, base/head SHAs, and severity triage of feedback.

Source: https://github.com/obra/superpowers @ 3dcbd5c4b4 · License: MIT · Author: Jesse Vincent (obra)
Receipts: old · new · drift

	`claude-sonnet-4-6` (old)	`claude-sonnet-5` (new)
with_skill (mean ± sd)	0.668 ± 0.249	0.719 ± 0.226
baseline (mean ± sd)	0.540 ± 0.262	0.617 ± 0.200

case	old (mean ± sd)	new (mean ± sd)	Δ	verdict
`request-package-basic`	0.898 ± 0.030	0.866 ± 0.021	-0.032	within noise
`full-handoff-before-merge`	0.400 ± 0.000	0.396 ± 0.009	-0.004	within noise
`mandatory-vs-optional-triggers`	0.862 ± 0.013	0.864 ± 0.013	+0.002	within noise
`resist-simple-self-review`	0.864 ± 0.009	0.882 ± 0.022	+0.018	within noise
`identify-base-head-shas`	0.800 ± 0.000	0.824 ± 0.018	+0.024	within noise (below floor)
`crafted-context-not-session-history`	0.290 ± 0.022	0.384 ± 0.188	+0.094	within noise
`triage-review-findings`	0.560 ± 0.029	0.820 ± 0.109	+0.260	improved

`writing-plans` — Writing Plans MIXED

How to author a concrete, self-contained implementation plan before coding.

Source: https://github.com/obra/superpowers @ 3dcbd5c4b4 · License: MIT · Author: Jesse Vincent (obra)
Receipts: old · new · drift

	`claude-sonnet-4-6` (old)	`claude-sonnet-5` (new)
with_skill (mean ± sd)	0.811 ± 0.089	0.785 ± 0.145
baseline (mean ± sd)	0.576 ± 0.230	0.675 ± 0.199

case	old (mean ± sd)	new (mean ± sd)	Δ	verdict
`task-right-sizing-testable-deliverable`	0.816 ± 0.094	0.460 ± 0.150	-0.356	regressed
`self-review-spec-coverage`	0.882 ± 0.011	0.844 ± 0.029	-0.038	within noise
`full-small-plan-header-and-tasks`	0.882 ± 0.030	0.850 ± 0.041	-0.032	within noise
`bite-sized-tdd-steps`	0.864 ± 0.013	0.876 ± 0.005	+0.012	within noise
`file-structure-by-responsibility`	0.802 ± 0.004	0.826 ± 0.037	+0.024	within noise
`repair-placeholder-steps`	0.802 ± 0.004	0.836 ± 0.035	+0.034	within noise
`interfaces-exact-signatures`	0.626 ± 0.150	0.804 ± 0.009	+0.178	improved

`skill-creator` — Skill Creator IMPROVED (1)

Conventions for authoring a SKILL.md: description-as-trigger, progressive disclosure, instructional tone.

Source: https://github.com/anthropics/skills @ b29e7cf65e · License: Apache-2.0 · Author: Anthropic, PBC
Receipts: old · new · drift

	`claude-sonnet-4-6` (old)	`claude-sonnet-5` (new)
with_skill (mean ± sd)	0.842 ± 0.095	0.875 ± 0.012
baseline (mean ± sd)	0.828 ± 0.070	0.774 ± 0.115

case	old (mean ± sd)	new (mean ± sd)	Δ	verdict
`restructure-oversized-skill`	0.882 ± 0.018	0.854 ± 0.032	-0.028	within noise
`reframe-rigid-musts`	0.876 ± 0.005	0.872 ± 0.004	-0.004	within noise
`domain-variant-organization`	0.876 ± 0.015	0.872 ± 0.008	-0.004	within noise
`write-triggering-description`	0.874 ± 0.013	0.874 ± 0.011	+0.000	within noise
`critique-and-fix-skillmd`	0.890 ± 0.035	0.894 ± 0.025	+0.004	within noise
`output-format-template`	0.868 ± 0.011	0.880 ± 0.012	+0.012	within noise
`draft-full-skillmd`	0.628 ± 0.074	0.882 ± 0.022	+0.254	improved

Methodology

Sampling & effect floor. Each (case, mode) generation is judged 5 times; the per-case band is mean ± sample stddev over those scores. A regression/improvement is claimed only when two with_skill bands are fully separated and the mean moved by at least the 0.05 effect floor. The floor exists because the judge quantizes to a coarse ~0.05–0.1 grid, so a confident grade often collapses to a zero-width “point band”; without the floor, two point bands one quantum apart (e.g. 0.60 vs 0.64) would read as a verdict despite no meaningful change.
Judge. claude-haiku-4-5, sampling mode surface-controlled. On the api surface the judge is pinned to temperature 0; this report ran on the claude-cli surface, where sampling is surface-controlled and cannot be pinned (recorded honestly in run.judge). That residual judge variance is exactly why verdicts require band separation, not a raw delta.
Text-only scoping. Report #001 tests skills whose value shows up in text (review, commit, documentation, prose, naming, planning). Tool-execution skills (document renderers, diagram/asset generators, browser/CLI tools) are out of scope — future work, needing a sandboxed-execution harness. Each skill is supplied as its single SKILL.md; bundled reference files are out of scope, and skills whose value lives entirely in bundled files were excluded.
Suite authorship (disclosure). We wrote the eval suites, one per skill, each case grounded in a claim documented in that skill's own SKILL.md. Rubrics anchor a fully-correct answer at 0.80 so a competent run is not saturated and a regression can show. Third-party skill content is not committed: each SKILL.md is fetched at run time from its pinned commit and sha256-verified.
Grounding policy (codified). Every gradable rubric criterion must trace to text present in the skill's SKILL.md at the pinned SHA. Criteria that import domain knowledge the skill does not state are not gradable, regardless of how standard that knowledge is. Claims may summarize; rubrics may not extrapolate. Each case records a grounding reference; the gate requires it. v1.2 applied this policy across all 10 suites (sweep) — see Amendments.
Baseline. Same prompt, no SKILL.md.

Reproduce

node scripts/fetch-skills.js
node scripts/run-report-001.js --concurrency 5
node scripts/build-report-001.js

Amendments

v1.2 · 2026-07-30 — In-text grounding policy adopted and applied across all 10 suites. Headline: v1.1 read 3 regressed / 3 improved / 3 mixed / 1 within noise; v1.2 reads 2 regressed / 3 improved / 4 mixed / 1 within noise. This supersedes v1.1.

Policy adopted (codified, applied unilaterally). Every gradable rubric criterion must trace to text present in the skill's SKILL.md at the pinned SHA. Criteria that import domain knowledge the skill does not state are not gradable, regardless of how standard that knowledge is. Claims may summarize; rubrics may not extrapolate. The grading standard does not depend on the audited party's tolerance. Full sweep of all 10 suites (70 cases): reports/rubric-sweep.md.
Criteria amended (2 of 10 suites; the other 8 were grounded and unchanged). git-workflow-and-versioning/semver-clean-bump required "the highest-precedence bump governs when changes are combined / reset PATCH to 0" — a rule the SKILL.md never states (it defines only MAJOR/MINOR/PATCH). That criterion was removed; the case now grades only the stated MAJOR/MINOR/PATCH definitions. commit-work (all 7 cases) required a specific Conventional Commits type token (e.g. chore(deps) vs feat) and, in one case, a subject <=72-chars / no-trailing-period / imperative-mood rule — none of which its SKILL.md states (it names Conventional Commits and shows the type(scope): summary shape but does not enumerate the type vocabulary). Type grading is now shape-only; the length/period/mood rule was removed. The logical-split, patch-staging, and secret/debug checks the SKILL.md does state remain fully graded.
Re-run. Only the two amended suites were re-run (both models, n=5, claude-cli, transcripts retained); the amendment changes only those two suite_hashes. The grounding reference now recorded on every case is excluded from suite_hash, so annotating the other eight suites did not disturb their receipts.
Verdict changes (stated plainly, not softened). The amended case git-workflow-and-versioning/semver-clean-bump flips regression → within noise (v1.1 Δ−0.054 on separated bands; v1.2 Δ−0.004). The v1.1 "regression" was driven by the imported precedence/reset criterion; grading only the MAJOR/MINOR/PATCH definitions the SKILL.md actually states, the new model shows no drift on this case. But re-running a suite re-rolls all seven of its cases (one receipt covers the whole suite), and the fresh n=5 measurement surfaced, on two unamended, well-grounded cases, a real regression and a real improvement: commit-message-conventional-type now regresses (Δ−0.256 — the new model wrote feat: instead of fix: for a bug fix; all five judge samples scored 0.60 with that same reason, transcript-verified) and semver-hidden-breaking-change now improves (Δ+0.062). Net, the skill moves REGRESSED(1) → MIXED. commit-work is unchanged at IMPROVED(1): its driver (two-sentence-describability-test, Δ+0.168) held, and its amended cases (specific type token → shape, and the removed ≤72-char/no-period/imperative-mood rule) remain within noise. The other eight skills were not re-run and keep their v1.1 labels.

v1.1 · 2026-07-28 — Pre-launch QA corrections. Net headline change: v1.0 read 4 regressed / 3 improved / 3 mixed / 0 within noise; v1.1 reads 3 regressed / 3 improved / 3 mixed / 1 within noise. Two causes, below.

(1) Minimum-effect floor of 0.05. A case is now called regressed/improved only when its two with_skill bands do not overlap AND the mean moved ≥ 0.05 (one judge-quantization step) — band separation alone is no longer sufficient, because the judge quantizes to a coarse grid and a confident grade often collapses to a zero-width point band. This reclassified 3 hair-trigger cases (|Δ| 0.014–0.024) to “within noise (below effect floor)”: code-review-and-quality/security-axis-sql-injection (which flips that whole skill REGRESSED → WITHIN NOISE), naming-analyzer/misleading-name-mutation-js (MIXED 2r/3i → 2r/2i), and requesting-code-review/identify-base-head-shas (IMPROVED 2 → 1).
(2) Two artifact-contaminated receipts re-run. A transcript audit found new-model bands containing a judge parse-fallback score of exactly 0.0 (an unparseable judge reply defaults to 0, per lib/judge.js). Both __claude-sonnet-5 receipts were re-run on the same claude-cli surface, n=5 (old-model receipts, which were clean, were kept as-is): writing-clearly-and-concisely and documentation-and-adrs.
What the re-run showed. active-voice-logging-passage’s v1.0 “regression” was entirely the artifact — the clean re-run scores 0.862 (vs old 0.852), i.e. within noise. match-existing-adr-convention reconfirmed as a genuine new-model failure (~0.08 again on an independent run). Because a receipt covers all 7 of a skill’s cases, re-running also refreshed the other cases honestly: comment-intent-not-implementation’s v1.0 regression did not reproduce (now within noise), while two different, real regressions surfaced — documentation-and-adrs/surface-conflicting-adr-conventions (Δ−0.268) and writing-clearly-and-concisely/concrete-language-incident-summary (Δ−0.152). Both skills keep their v1.0 labels (MIXED and REGRESSED-1 respectively); only the driving cases changed.
Full per-case transcript evidence and REAL-vs-artifact classifications are in reports/report-001-audit.md.

Receipts validate against the Driftproof receipt spec (v0.2; the two suites re-run in v1.2 are v0.3, a superset the loader also accepts). Every verdict above is a function of the two with_skill bands recorded in the receipts. Verdicts age because the substrate moves — that is the phenomenon this report measures.

Report #001 — do skill verdicts hold across a model release?

Per-skill summary

code-review-and-quality — Code Review and Quality WITHIN NOISE

git-workflow-and-versioning — Git Workflow and Versioning MIXED

documentation-and-adrs — Documentation and ADRs MIXED

commit-work — Commit Work IMPROVED (1)

writing-clearly-and-concisely — Writing Clearly and Concisely REGRESSED (1)

crafting-effective-readmes — Crafting Effective READMEs REGRESSED (2)

naming-analyzer — Naming Analyzer MIXED

requesting-code-review — Requesting Code Review IMPROVED (1)

writing-plans — Writing Plans MIXED

skill-creator — Skill Creator IMPROVED (1)

Methodology

Reproduce

Amendments

`code-review-and-quality` — Code Review and Quality WITHIN NOISE

`git-workflow-and-versioning` — Git Workflow and Versioning MIXED

`documentation-and-adrs` — Documentation and ADRs MIXED

`commit-work` — Commit Work IMPROVED (1)

`writing-clearly-and-concisely` — Writing Clearly and Concisely REGRESSED (1)

`crafting-effective-readmes` — Crafting Effective READMEs REGRESSED (2)

`naming-analyzer` — Naming Analyzer MIXED

`requesting-code-review` — Requesting Code Review IMPROVED (1)

`writing-plans` — Writing Plans MIXED

`skill-creator` — Skill Creator IMPROVED (1)