From b412c90d05706b5a1ab750356c494007063039a6 Mon Sep 17 00:00:00 2001 From: jprisant <6282841+jprisant@users.noreply.github.com> Date: Sat, 12 Sep 2026 00:31:58 -0700 Subject: [PATCH] chore(marketplace): re-pin thinking-framework-skills to v0.15.0 Installers get the new build only after this merges. sha e317da9 -> 5e44707 (the v0.15.0 tag commit) version 0.14.0 -> 0.15.0 metadata 1.77.0 -> 1.78.0 ## What v0.15.0 contains No new skills; the catalog stays 56 evidence-graded core skills + 7 contested lenses. 13 PRs (#117-#129) in the upstream repo. **Everything the library publishes about itself is now either machine-asserted or measured by an instrument that can actually see it.** Four behavioural runs closed the last measurement gaps: - All four entry-point tools reach `maturity: measured`. The fourth needed a purpose-built dispatch harness, because it delegates to a subagent and its own instructions forbid running research inline - so the generic harness could only ever have graded it breaking its own procedure. - The front-door question ("does an agent with 77 tools installed reach for this library at all?") is measured and published as a distinct third question, never blended with the other two. - The catalog-enrichment question is answered by a controlled single-variable A/B. - The 63-skill artifact-quality figure was re-measured and **went down**, 428/429 -> 424/429, because the grader was rewritten to stop believing artifacts' claims about other skills. Published as-is, with the reason, rather than smoothed. **Four defects the runs found, all fixed:** a score generator that rounded 99.5%+ up to a flat `100%` (the cause of a false claim on the public trust page), a skill publishing the wrong evidence tier for three months, a skill telling users to hand-roll a method this library ships, and the rounding's knock-on effect of hiding a real false-fire result. **Contributor road:** `npm run setup` takes a fresh clone to a green gate in one command, reading the pinned validator commit out of CI so it cannot drift from what CI grades against. Issue and PR templates encode the selection bar. Conformance gate stays fifteen layers, `0 errors / 90 warnings` at declared Standard 0.8 (the `G9` docblock debt was paid off in full, 129 -> 90). Test suite 286 -> 326. ## Verification - `sha` is the exact commit the `v0.15.0` tag points at (`git rev-parse v0.15.0`), not a branch head - `marketplace.json` re-parsed as valid JSON after editing - Upstream release: https://github.com/product-on-purpose/thinking-framework-skills/releases/tag/v0.15.0 Co-Authored-By: Claude Opus 5 Claude-Session: https://claude.ai/code/session_011eP4ewzw2X9FuPTo1hKUQ4 --- .claude-plugin/marketplace.json | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/.claude-plugin/marketplace.json b/.claude-plugin/marketplace.json index 198d3e6..c69ada2 100644 --- a/.claude-plugin/marketplace.json +++ b/.claude-plugin/marketplace.json @@ -7,7 +7,7 @@ }, "metadata": { "description": "Product on Purpose: thematic AI agent skill and tool collections for product work.", - "version": "1.77.0" + "version": "1.78.0" }, "plugins": [ { @@ -26,10 +26,10 @@ "source": { "source": "url", "url": "https://github.com/product-on-purpose/thinking-framework-skills.git", - "sha": "e317da999be8c7c29d0da4cdb98d2fdd3e4a1374" + "sha": "5e44707956b746360512bc814ac924f5225bb472" }, "description": "Evidence-graded, agent-executable thinking-method skills: each is reduced to its working mechanism, graded honestly on its evidence, and produces a concrete artifact. Includes a published Framework Library, a research engine, and routing/applicator tools. Follows the agentskills.io specification.", - "version": "0.14.0", + "version": "0.15.0", "strict": true }, {