CLAUDE.md: agent 4-eyes before every commit - #482
Merged
Merged
Conversation
Codifies the session directive that caught a CSS-breaking template defect, a fabricated quote on a live page, and 3 orphaned migration references - all after automated gates were already green. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Contributor
|
Important
This repository does not receive automatic reviews because it has fewer than 10 stars. ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Plus Run ID: Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
pftg
added a commit
that referenced
this pull request
Aug 20, 2026
#483) * Course/blog: dense tables scroll inside themselves on phones (C3.1) W2's scroll gate flagged this in August and the fix was frozen 'until this wave'. Reproduced before fixing, and it was worse than recorded: on fractional-cto-sow-reference a table rendered 507px wide inside a 354px column (recorded as 25px), with overflow-x visible on the parent - no scroll container anywhere. first-paying-customer-operating-kit overflowed 51px, pre-existing since W1 #428. Mobile-only media query in single-post.css, which both single.html and course/single.html already load, so one rule covers blog and course and every sibling reference table at once. display:block is what makes a table scrollable; scoping it under 767px keeps the desktop auto-layout the wide reference tables depend on. Verified at a true 390x844 device render: all four tables now cap at the 354px column and scroll internally, document scrollWidth == viewport on both named pages (was overflowing). Desktop re-checked at 1280: display:table, overflow-x:visible, full 684px column - untouched. bin/qtest --changed green (34 runs, 53 screenshots, 0 failures). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Purge fabricated testimonials from four live pages; label invented case studies Found by following C0.1's ratchet output: 25 of its 40 violations traced to client copy syndicated across every services page. Pulling that thread surfaced something much worse than a banned adjective. **data/testimonials.yaml carried four synthetic testimonials** - Sarah Chen / TechStart Solutions, David Rodriguez / EcomMega, Maria Gonzalez / HealthTech Innovations, Alex Thompson / EduPlatform - with placeholder-shaped company names and unsourced round metrics ('Secured $2M Series A', '300% improvement in page load speeds', '99.8% uptime'). They rendered on the LIVE homepage, /clients, /about-us and /use-cases, because this data file overrides the curated single testimonial in hugo.toml. Someone had already curated hugo.toml down to the one real testimonial (Bruno Wozniak / PubNative, corroborated by a live LinkedIn profile and the Review schema). The data file quietly overrode that curation in the visible render. Same defect class the schema purges of 2026-08-14 and #475 removed from structured data - this was the visible-copy half nobody had swept. Removed all four. Kept Bruno's verbatim quote and attribution, dropped his block: three round percentages that appear nowhere in his quote and trace to no source. Restore path is in the file header. Also: fractional-cto-roi-calculator presented five full-name founders with direct quotes under 'Real Founder Stories ... from JetThoughts clients', with no anonymisation disclaimer - and said 'three stories' while showing five. Its sibling infrastructure-spending post already carries the honest form ('names changed, numbers accurate'). Relabelled as composites with the count fixed. NOTE: 'Sarah Chen' appears both here and in the synthetic testimonials, which is why I did not simply assert these are real engagements - see the handback. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * OKF: session wrap-up - testimonial canon rule, estimation lesson, two coverage gaps Four durable captures from the 2026-08-20 session that were not already logged as they happened: - claims-canon gains a testimonial rule. The finding worth keeping is not 'four fake testimonials existed' but the MECHANISM: hugo.toml had been curated to the one real testimonial, and data/testimonials.yaml silently outranked it because the partial reads the data file first. Cleaning a config is not cleaning the page - check which source wins. - Query before estimating. A panel guessed a click rate GA could answer, and a kill-criterion override shipped on the guess at ~5x off, which would have made the campaign unkillable. Records the GA-vs-LinkedIn division of labour so it stops drifting back onto Paul. - The visual suite does not cover the testimonial carousel: four live pages changed, zero of 53 screenshots moved. - Fresh agent worktrees need bun install v1.3.13 (bf2e2cec) + cssnano@8.0.6 + postcss@8.5.26 + postcss-import@17.0.0 + postcss-nested@8.0.1 + surge@0.43.1 + caniuse-lite@1.0.30001809 51 packages installed [518.00ms] before bin/dev. Not re-logged: the exhibit two-layer trap, the fallback-font em correction, the ratchet-reads-built-HTML change and the analytics read were all written to .okf/log.md as they happened. The 4-eyes rule another session added to CLAUDE.md (#482) already covers 'self-review is not the second pair of eyes', so this session's identical finding is corroboration, not a new rule. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
pftg
added a commit
that referenced
this pull request
Aug 20, 2026
* Course/blog: dense tables scroll inside themselves on phones (C3.1) W2's scroll gate flagged this in August and the fix was frozen 'until this wave'. Reproduced before fixing, and it was worse than recorded: on fractional-cto-sow-reference a table rendered 507px wide inside a 354px column (recorded as 25px), with overflow-x visible on the parent - no scroll container anywhere. first-paying-customer-operating-kit overflowed 51px, pre-existing since W1 #428. Mobile-only media query in single-post.css, which both single.html and course/single.html already load, so one rule covers blog and course and every sibling reference table at once. display:block is what makes a table scrollable; scoping it under 767px keeps the desktop auto-layout the wide reference tables depend on. Verified at a true 390x844 device render: all four tables now cap at the 354px column and scroll internally, document scrollWidth == viewport on both named pages (was overflowing). Desktop re-checked at 1280: display:table, overflow-x:visible, full 684px column - untouched. bin/qtest --changed green (34 runs, 53 screenshots, 0 failures). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Purge fabricated testimonials from four live pages; label invented case studies Found by following C0.1's ratchet output: 25 of its 40 violations traced to client copy syndicated across every services page. Pulling that thread surfaced something much worse than a banned adjective. **data/testimonials.yaml carried four synthetic testimonials** - Sarah Chen / TechStart Solutions, David Rodriguez / EcomMega, Maria Gonzalez / HealthTech Innovations, Alex Thompson / EduPlatform - with placeholder-shaped company names and unsourced round metrics ('Secured $2M Series A', '300% improvement in page load speeds', '99.8% uptime'). They rendered on the LIVE homepage, /clients, /about-us and /use-cases, because this data file overrides the curated single testimonial in hugo.toml. Someone had already curated hugo.toml down to the one real testimonial (Bruno Wozniak / PubNative, corroborated by a live LinkedIn profile and the Review schema). The data file quietly overrode that curation in the visible render. Same defect class the schema purges of 2026-08-14 and #475 removed from structured data - this was the visible-copy half nobody had swept. Removed all four. Kept Bruno's verbatim quote and attribution, dropped his block: three round percentages that appear nowhere in his quote and trace to no source. Restore path is in the file header. Also: fractional-cto-roi-calculator presented five full-name founders with direct quotes under 'Real Founder Stories ... from JetThoughts clients', with no anonymisation disclaimer - and said 'three stories' while showing five. Its sibling infrastructure-spending post already carries the honest form ('names changed, numbers accurate'). Relabelled as composites with the count fixed. NOTE: 'Sarah Chen' appears both here and in the synthetic testimonials, which is why I did not simply assert these are real engagements - see the handback. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * OKF: session wrap-up - testimonial canon rule, estimation lesson, two coverage gaps Four durable captures from the 2026-08-20 session that were not already logged as they happened: - claims-canon gains a testimonial rule. The finding worth keeping is not 'four fake testimonials existed' but the MECHANISM: hugo.toml had been curated to the one real testimonial, and data/testimonials.yaml silently outranked it because the partial reads the data file first. Cleaning a config is not cleaning the page - check which source wins. - Query before estimating. A panel guessed a click rate GA could answer, and a kill-criterion override shipped on the guess at ~5x off, which would have made the campaign unkillable. Records the GA-vs-LinkedIn division of labour so it stops drifting back onto Paul. - The visual suite does not cover the testimonial carousel: four live pages changed, zero of 53 screenshots moved. - Fresh agent worktrees need bun install v1.3.13 (bf2e2cec) + cssnano@8.0.6 + postcss@8.5.26 + postcss-import@17.0.0 + postcss-nested@8.0.1 + surge@0.43.1 + caniuse-lite@1.0.30001809 51 packages installed [518.00ms] before bin/dev. Not re-logged: the exhibit two-layer trap, the fallback-font em correction, the ratchet-reads-built-HTML change and the analytics read were all written to .okf/log.md as they happened. The 4-eyes rule another session added to CLAUDE.md (#482) already covers 'self-review is not the second pair of eyes', so this session's identical finding is corroboration, not a new rule. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * Remove the fabricated founder stories (Paul: 'she is not real') Sarah Chen is confirmed invented. She appeared in both the synthetic testimonials purged in #481 and, with a surname and a direct quote, in a live post - so the whole batch goes. fractional-cto-roi-calculator: removed the five-founder section entirely (Sarah Chen, Marcus Thompson, Jennifer Martinez, David Park, Emma Rodriguez) along with BOTH of its false claims - the 'Real Founder Stories ... from JetThoughts clients' heading and the closing 'these aren't hypothetical scenarios - they're real situations from JetThoughts clients in the past 24 months'. The five underlying patterns are worth keeping, so they stay as an unattributed list that states plainly we have no completed case study of our own to publish yet. infrastructure-spending-evaluation: corrected in the same pass. Its 'real stories from our clients (names changed, numbers accurate)' form had looked like the honest convention - I nearly left it on that basis earlier today. With the fabrication next door confirmed, an unverifiable 'numbers accurate' claim is the same defect in better clothes. Rule recorded in claims-canon: a disclaimer that asserts MORE than you can verify is not a fix. 'Names changed, numbers accurate' claims real provenance. If nobody can produce the client, write 'illustrative composite'. Repo-wide sweep: zero surviving instances of the invented personas or the real-client claims outside the audit trail in data/testimonials.yaml. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Codifies Paul's 2026-08-20 directive ("spawn appropriate agents… and re-review each changes before commit") as a standing behavioral constraint: at least one reviewer agent on the actual diff before any non-trivial commit — correctness reviewer for code/templates, cold-eyes ICP reader for customer-facing copy; verdict quoted in the commit/PR.
Same-day evidence it pays: pre-commit reviewers caught a markdownify
<p>-stripping defect (would have broken CSS on every service page), a fabricated founder quote on the live rescue page, and 3 orphaned references in the ops migration — all after hugo-build/qtest were green.Proposed via /reflection, approved by Paul. Docs-only.
🤖 Generated with Claude Code