01 Data analytics
KPI CreatorAgent team · Ovy, Gary, Kewpie, Viktor, Jennifer
A five-agent team that turns a business problem into a tested KPI dashboard, stopping for your approval at every stage.
26Builds: 24 Claude Code, 2 ChatGPT
~50h 15mClaude Code build time (ChatGPT not recorded)
~2h 06mAverage per Claude Code build
20 vs 2Success vs failures: running vs issues being addressed (3 in development, 1 not set)
92 · ~9h 35mClaude Code uses and total run time
Status
Completed & running 20 In development 3 Issues being addressed 2 Not set 1Status as at 29 Oct 2026, set by Adam.
Build timeline
Running total of builds. Labels show builds completed each month: 4 + 1 + 7 + 14 = 26. Hover a point to see what was built that day.
Month by month
- Oct+14 builds so far, 26 in total, including the Dragonboat Seat Planner (~8h 30m), the KPI Creator agent team (~7h 50m) and Cardio Story (~5h 25m).
- Sep+7 builds, 12 in total: 2 ChatGPT job tools, the company research suite and 2 GitHub skills.
- Aug+1 build, 5 in total: Home Internet Speed Report, tracking home internet speed by time of day.
- Jul4 builds in the first 3 days, including the Portfolio Website (~11h 15m), the longest build.
Claude Code build time by category
Select a category to see its builds. Numbers show builds per category.
Skills and tools used
ClaudeChatGPTGitHubApps and servicesCode and formatsResearch and inputs
40 skills and tools across 26 builds. Bubble size shows how many builds use each one.
Select a bubble to see which builds use it. The bubbles grow as the build timeline plays.
Top 3 Highlights
02 Sport & community
Dragonboat Seat PlannerWeb app for coaches
Balances a 20-paddler dragon boat crew by weight, experience and role in under a second, right on the coach's phone.
03 Career & job search
Company Research AgentAgent · "Company Briefs"
Researches a company's history, strategy, pay and why the role exists now, then delivers a staged interview brief and web report.
Builds by month
July 4
- Portfolio WebsitePersonal brand site: career story, stats and certifications.Built26 Jul to 27 OctBuild time~11h 15mUsed82 page views
- Daily Job ScannerDaily scan of job sites for senior data roles, rated Tier 1 or 2.Built26 JulBuild time~1h 10mUsed15 uses
- Publish to GitHubPublishes any finished HTML file to GitHub Pages.Built27 JulBuild time~30mUsed1 use
- What's On TodayMorning brief plus today's calendar.Built28 JulBuild time~15mUsed2 uses
August 1
- Home Internet Speed ReportSummarises home internet speed tests by time of day.Built19 AugBuild time~35mUsed9 uses
September 7
- Morning/Evening Job Search ChatGPT6am briefing and 5pm review of fresh roles, ranked by fit.Built1 SepBuild timeNot recordedUsed~95 uses
- Company Research Plugin ChatGPTCompany research and interview dossiers, now at v0.3.2.Built25 SepBuild timeNot recordedUsedNot recorded
- Company Research AgentRuns the interview brief, then hands off to the web report.Built28 SepBuild time~20mUsed4 uses
- Build Company BriefCompany history, strategy, pay and why the role exists now.Built28 SepBuild time~45mUsed4 briefs
- Build Company Brief HTML ReportTurns an approved brief into a polished HTML report.Built28 SepBuild time~45mUsed5 reports
- New GitHub SiteCreates a new GitHub Pages site and puts a report live.Built29 SepBuild time~10mUsed8 uses
- Update Live SitePushes the latest report to its live site.Built29 SepBuild time~10mUsed15 uses
October 14
- ACCA Event Info SheetOne-page A4 PDF for ACCA race days and events.Built10 OctBuild time~35mUsed0 uses
- Dragonboat Seat PlannerBalances a 20-seat dragon boat crew in under a second.Built12 OctBuild time~8h 30mUsedNot tracked
- Junior Touch Footy Attendance PollPrepares the weekly game poll from the Norths email.Built13 OctBuild time~1h 30mUsed12 uses
- AI Build LogThis report: every build with dates, build time, usage and status.Built18 OctBuild time~5h 30mUsedNot tracked
- AI Brain GitHub SyncKeeps the AI Brain in a private GitHub repo, synced each session.Built18 OctBuild time~30mUsed79 commits
- UX Standards ReviewChecks a page against your UX standards at 5 screen sizes.Built19 OctBuild time~25mUsed~4 uses
- HTML Standards ReviewChecks HTML, accessibility, speed and security against your build standards.Built19 OctBuild time~25mUsed~4 uses
- Digital Pollen AnimationAdds the slow digital pollen animation behind a page section.Built19 OctBuild time~30mUsed3 uses
- AI Build Log UpdaterScans every build and lists log changes for approval.Built19 OctBuild time~25mUsed3 uses
- KPI CreatorAgent team that turns a business problem into a KPI dashboard.Built20 OctBuild time~7h 50mUsed3 test runs, 1 live run
- Agent SimulationAnimated replay of a KPI Creator run, agent by agent.Built23 OctBuild time~1h 30mUsed2 uses
- KPI Token ReportToken use per KPI Creator run, by agent or side by side.Built24 OctBuild time~45mUsedNot yet by command
- KPI End-to-End UpdaterAdds a finished test case to the public end-to-end page.Built27 OctBuild time~30mUsedNot yet
- Cardio StoryHow VO2 Max rose with dragon boating, from Apple Health data.Built28 OctBuild time~5h 25mUsedNot tracked
All builds
Select a build to see its details.
Portfolio WebsiteGitHub Pages site
Successfully Completed & Running
Claude Code
- Developed
- 26 Jul 2026Live 30 JulLast update 27 Oct
- Build time
- ~11h 15mOver 15 sessions, 26 Jul to 20 Oct
- Times used
- 82page views since 14 Aug
- Total run time
- Always onWebsite, no run time
- Version updates
- 45Published updates, 30 Jul to 27 Oct
What it does
- Shows your career track, headline stats, capability stories, skills and certifications
- Resume links served from the GitHub repo
- Links to the Data Strategy & Evolution page
- Visits tracked with GoatCounter
Results
- Live at adamaziz1979.github.io/AdamAziz_DataProfessional
- 45 updates published, 30 Jul to 27 Oct
- AI Builds section, 1-minute video and LinkedIn share thumbnail added 19 Oct
- Used as evidence in every Build Company Brief
Obstacles
- First upload was missing its CSS, script and resume files; now built into one index.html
- Timeline section removed on 2 Aug until it is ready
- Visit alert routine built 14 Aug, switched off since 15 Aug
Skills and tools
Claude CodeHTML, CSS, JSGitHub PagesGitHub CLIGoatCounter
AI Build LogReport · artifact + GitHub Pages
Successfully Completed & Running
Claude Code
- Developed
- 18 Oct 2026Live 18 OctLast update 27 Oct
- Build time
- ~5h 30mOver 4 sessions, 18 to 26 Oct (estimate)
- Times used
- Not trackedNo visit counter yet
- Total run time
- Always onWeb page, no run time
- Version updates
- 34Public releases, 18 to 27 Oct
What it does
- Tracks every AI build with dates, build time, usage, status, results and obstacles
- Statuses set in dropdowns and saved with the page
- Animated KPIs, build timeline with month summaries and category bars
- Dark public copy on GitHub Pages
Results
- Live at adamaziz1979.github.io/AI_Build_Log
- 26 builds logged across Claude Code and ChatGPT
- Statuses saved per build and carried into the public page
- Top 3 Highlights carousel and nested skills added 26 Oct
Obstacles
- Timeline labels read as 12 builds, not 16; relabelled by month so they add up
- Summary list showed in full on GitHub Pages; fixed the hiding rule
- Animations stalled in background tabs; added a timer fallback
Skills and tools
Claude CodeClaude artifactArtifact databaseHTML, CSS, JSSVG animationGitHub PagesGitHub CLI
Daily Job ScannerSkill · /job-scanner
Issues being addressed
Claude Code
- Developed
- 26 Jul 2026Live 28 Jul
- Build time
- ~1h 10mOver 3 days, incl. Chrome fixes+~25m refinement 10 Oct
- Times used
- 1512 scheduled, 3 manual
- Total run time
- ~2h 05mClaude working time
- Version updates
- 227 and 28 Jul
What it does
- Scans SEEK, Michael Page, LinkedIn, Indeed, Employment Hero, Greenhouse and other job sites daily
- Only roles posted in the last 24 hours or 7 days
- Labels each role new, refreshed or still active, then rates it Tier 1 or 2
- Logs every role in tracker.csv (156 seen so far)
Results
- 12 scan reports, 27 Jul to 10 Oct
- 156 roles tracked, deduplicated across sites
Obstacles
- Claude in Chrome permission bug in Jul; added a web search fallback
- Michael Page dashboard is login-only, so it gets skipped in fallback mode
- Old cached listings showed up; limited results to the last 24 hours and 7 days
- Daily schedule is now switched off (last run 14 Oct)
Skills and tools
Claude Code skillClaude in ChromeWeb searchScheduled taskCSV tracker
Morning/Evening Job Search2 ChatGPT scheduled tasks · 6am briefing and 5pm review
Successfully Completed & Running
ChatGPT
- Developed
- 1 Sep 2026Chats created 6:03am and 4:58pmScheduler dates unknown
- Build time
- Not recordedRefined Sep to Oct
- Times used
- ~952 runs a day since 1 Sep, not verified
- Total run time
- Not recorded
- Version updates
- Not recorded5pm task now on V2
What it does
- Briefs you at 6am on fresh roles, recruiter email updates and priority actions
- Reviews permanent and contract roles at 5pm, ranked by fit
- Only roles posted in the last 1 to 2 days
- Tracks application status and pipeline changes
Results
- Daily outputs through 18 Oct
- 17 Oct: 5 fresh roles at 6am; 1 provisional permanent role at 5pm
- 18 Oct: 1 application acknowledged, 1 rejection
- Labels conflicting ad dates instead of over-claiming
Obstacles
- Old or refreshed ads shown as new; now checks original posting dates
- Posting dates conflict across job sites
- Strict criteria give far fewer than the target 10 roles per lane
- The two tasks overlap; scheduler status not verified
Skills and tools
ChatGPT scheduled taskGmailWeb search
Company Research PluginPlugin skill · "Company Research & Brief"
Successfully Completed & Running
ChatGPT
- Developed
- 25 Sep 2026v0.3.1; v0.3.2 on 28 SepWorkflow may be older
- Build time
- Not recordedv0.3.2 update: ~2m agent run
- Times used
- Not recorded
- Total run time
- Not recorded
- Version updates
- 1v0.3.1 to v0.3.2
What it does
- Researches companies and writes executive briefings and interview dossiers tailored to role and candidate
- Two-stage briefing with explicit evidence classifications
- Since v0.3.2: pay, headcount, market benchmarks and salary-expectation wording
Results
- Reusable installed skill, now at v0.3.2
- Includes 2 reference reports
- Became the base for Build Company Brief in Claude Code (28 Sep)
Obstacles
- Local headcount and exact-role pay often not public; handled with dated sources, confidence labels and "not publicly disclosed"
- Must separate fact from inference and use the latest resume
- HTML needs a separate approval step
- Time saved and interview outcomes not measured
Skills and tools
ChatGPT plugin skillWeb searchHTML report
Company Research AgentAgent · "Company Briefs"
Successfully Completed & Running
Claude Code
- Developed
- 28 Sep 2026
- Build time
- ~20mSame night as the brief skill
- Times used
- 5Same uses as Build Company Brief
- Total run time
- IncludedCounted under the brief
- Version updates
- 129 Sep
What it does
- Prepares you for interviews when you call it by name
- Follows Build Company Brief, then hands off to Build Company Brief HTML Report
- Saves to OneDrive\Documents\Career\[Company]
- Never publishes unless you say "create repo" or "Update Github"
Results
- Called by name for every interview brief since 28 Sep
- Saves straight to the Career folder
Obstacles
- Agents cannot be opened from the sidebar; you call them by name in chat
- Output folder and resume check had to be added after the first run
Skills and tools
Claude Code agent/build-company-brief/build-html-reportOneDrive
Build Company BriefSkill · /build-company-brief
Successfully Completed & Running
Claude CodePorted from the ChatGPT plugin
- Developed
- 28 Sep 2026Final version 29 Sep
- Build time
- ~45mBuilt from plugin instructions
- Times used
- 55 different companies
- Total run time
- ~1h 35mClaude working time
- Version updates
- 2Both on 29 Sep
What it does
- Researches company history, turning points, strategy, headcount and pay
- Works out why the role exists now
- Checks your Master Resume, STAR Skills, portfolio and LinkedIn
- Delivers a staged executive briefing, dossier and Stage 3 focus points
Results
- 5 interview briefs, each for a different company
- Checks resume and STAR dates every time
Obstacles
- Reference links moved from each major section down to each sub-section
- Stage 3 (top 5 experience points) added after the first draft
Skills and tools
Claude Code skillWeb searchMaster ResumeSTAR SkillsLinkedIn
Build Company Brief HTML ReportSkill · /build-html-report
Successfully Completed & Running
Claude Code
- Developed
- 28 Sep 2026Final version 29 Sep
- Build time
- ~45mRefined on the first brief
- Times used
- 5Plus regenerations in brief sessions
- Total run time
- ~40m
- Version updates
- 229 Sep and 26 Oct
What it does
- Turns an approved brief into one self-contained HTML report
- Uses one approved template: executive infographic, share-price chart with 5+ event labels, Lexend font
- Can regenerate an existing report under the latest rules
Results
- 4 HTML reports from one approved template
- Executive infographic and labelled share-price chart in every report
- Retitled Company, Role and Recruitment Brief on 26 Oct; 4 live briefs updated
Obstacles
- Original font was hard to scan; switched to Lexend after comparing 8 options
- Share-price chart needed at least 5 labelled turning points
- Company website link added top-right after review
Skills and tools
Claude Code skillHTML, CSSSVG chartsLexend font
Junior Touch Footy Attendance PollSkill + 2 routines
Successfully Completed & Running
Claude Code
- Developed
- 13 Oct 2026Final version 21 Oct
- Build time
- ~1h 30mOver 3 days, incl. cloud routine
- Times used
- 124 on PC, 8 cloud, to 28 Oct
- Total run time
- ~30mCloud runs ~30s each
- Version updates
- 414, 16, 20 and 21 Oct
What it does
- Reads this week's U8 game from the Norths Touch round email
- Handles byes, missing emails and Gmail failures
- Prepares the WhatsApp attendance poll; posts only after you say yes
- Runs Wed 11am on your PC, with a cloud backup routine (repo touchfooty)
Results
- First poll posted 14 Oct (Round 4)
- Scheduled run correctly skipped a duplicate poll
- PC and cloud runs both succeeded on 21 Oct
Obstacles
- WhatsApp desktop app and Edge cannot be driven; it only works in Chrome
- PC routine needs the PC on with Claude and Chrome open
- Cloud routine cannot post to WhatsApp, so it sends the poll text to chat
- "Allow multiple answers" had to be turned off
Skills and tools
Claude Code skillGmailClaude in ChromeWhatsApp WebScheduled taskCloud routineGitHub
ACCA Event Info SheetSkill · /create_acca_infosheet
Successfully Completed & Running
Claude Code
- Developed
- 10 Oct 2026
- Build time
- ~35mFirst sheet, then saved as a skill
- Times used
- 01 sheet made while building it
- Total run time
- None yet
- Version updates
- 112 Oct
What it does
- Creates a one-page A4 PDF for ACCA events in the accadbr.com style
- You pick 3 to 5 sections: race schedule, seating, weather, logistics and more
- Pulls from race draws, crew lists and briefing files
Results
- DBNSW Round 1 info sheet (10 Oct), PDF and HTML
- Matches the accadbr.com style
Obstacles
- Team App auto-login not possible: the assistant cannot enter your passwords
- Mobile HTML version dropped in favour of a PDF
- Race table needed realigning against the bullet panel
Skills and tools
Claude Code skillPythonHTML, CSSHeadless Chrome PDF
Dragonboat Seat PlannerWeb app for coaches
Successfully Completed & Running
claude.ai chatv1 to v17Claude Codev18 to v72
- Developed
- 12 Oct 2026Live 12 Oct (v18 to v72)LinkedIn postPublic demo 16 Oct
- Build time
- ~8h 30mClaude Code, 12 to 19 OctEarlier claude.ai chat not counted
- Times used
- Not trackedRuns in coaches' browsers, no analytics
- Total run time
- Under 1 secPer plan: best of 8 runs, ~136,000 seatings checked
- Version updates
- 71v1 to v17 in a claude.ai chat, v18 to v72 in Claude Code; 28 pushed live
What it does
- Builds a balanced seating plan for a 20-seat dragon boat: left/right weight, front/back balance, experience and roles
- Roster, attendance, seat plan and share tabs; swap, lock and re-balance seats
- Shares a names-only boat image with the crew
- Installs on a phone and works offline at the dock
Results
- Coach app live at v72 after 28 updates, incl. a How to use flow (19 Oct)
- Most experienced paddlers always seated in the front rows in testing; left/right within 1 to 3 kg
- Frozen public demo shared in a LinkedIn post
- Roster CSV template for coaches
Obstacles
- Stack Team App has no public API, so attendance is ticked by hand
- BOM terms block reusing forecasts, so the app links to BOM instead
- Lone paddlers in rows 9 and 10 slipped past the weight-partner rule; fixed with a half-row penalty
- Coach app and demo share one github.io address and clear each other's offline cache; fix due in the next update
- Needs to move to Cloudflare so only approved coaches can open it
Skills and tools
claude.ai chatClaude CodeHTML, JS web appGitHub Pages
Publish to GitHubSkill · /publish-to-github
Successfully Completed & Running
Claude Code
- Developed
- 27 Jul 2026
- Build time
- ~30mIncl. GitHub CLI sign-in
- Times used
- 1Later publishing used New GitHub Site and Update Live Site
- Total run time
- ~20m
- Version updates
- 129 Sep
What it does
- Publishes any finished HTML file to GitHub Pages
- Asks new repo or existing, then does the rest
Results
- AI Brain Architecture page live on 27 Jul
- Base for the two GitHub skills that followed
Obstacles
- GitHub CLI needed a one-off sign-in before the first push
Skills and tools
Claude Code skillGitHub CLIGitHub Pages
New GitHub SiteSkill · /create-repo
Successfully Completed & Running
Claude Code
- Developed
- 29 Sep 2026
- Build time
- ~10mBuilt with Update Live Site
- Times used
- 8Latest: Cardio Story site, 29 Oct
- Total run time
- ~30m
- Version updates
- 0Unchanged since 29 Sep
What it does
- Creates a new public repo on adamaziz1979
- Pushes the report as index.html and turns on Pages
- Checks for existing repos for the same company first
Results
- 9 repos created since 29 Sep, including touchfooty and internet_speed_report
- Pages turned on automatically
Obstacles
- First push failed until GitHub Pages was enabled; now part of the skill
- Two repos for the same company caused confusion; it now checks first
Skills and tools
Claude Code skillGitHub CLIGitHub Pages
Update Live SiteSkill · /update-github
Successfully Completed & Running
Claude Code
- Developed
- 29 Sep 2026
- Build time
- ~10mBuilt with New GitHub Site
- Times used
- 15By command; plain "Update Github" requests not counted
- Total run time
- ~55m
- Version updates
- 0Unchanged since 29 Sep
What it does
- Replaces index.html in an existing repo
- Confirms the live page rebuilt
- Asks which repo when more than one matches
Results
- Main tool for updating live briefs, heavy use 12 to 13 Oct
- Confirms the live page rebuilt
Obstacles
- Some companies have more than one repo; it now asks instead of guessing
Skills and tools
Claude Code skillGitHub CLIGitHub Pages
Home Internet Speed ReportSkill · /speed-report
Issues being addressed
Claude Code
- Developed
- 19 Aug 2026Updated 14 Oct
- Build time
- ~35m15m build + 20m updateCollector built separately
- Times used
- 9
- Total run time
- ~15m
- Version updates
- 220 Aug and 14 Oct
What it does
- Summarises speed tests run 4 times a day (9am, 12pm, 3pm, 7pm)
- Averages by time slot, drops bad pings, flags failed runs
- Sydney servers only since 14 Oct
- Matches the live report on GitHub Pages
Results
- Live report at internet_speed_report on GitHub Pages
- Slot averages that match the published page
Obstacles
- Only 11 good readings logged since 19 Aug; 27 test samples failed
- Pinned server was dropped, so tests drifted to Melbourne, Perth and Brisbane
- Unreachable hosts returned 1.8M to 3.6M ms pings; now excluded
- Sydney-only rule skipped 5 runs on 14 to 15 Oct
- No new readings since 14 Oct
Skills and tools
Claude Code skillPythonspeedtest-cliWindows Task SchedulerGitHub Pages
What's On TodaySkill · /whats-on-today
In Development, Refinements Needed
Claude Code
- Developed
- 28 Jul 2026
- Build time
- ~15mSingle session
- Times used
- 2
- Total run time
- ~10m
- Version updates
- 0Unchanged since 28 Jul
What it does
- Runs the morning brief: inbox, top 3 priorities, loose ends
- Puts today's calendar up front
- Offers the rest of the week, flagging interviews
Results
- One command for today's brief and calendar
- Offers the rest of the week, flagging interviews
Obstacles
- No major obstacles
- Added a re-check so today's calendar never comes back empty
Skills and tools
Claude Code skillGmailGoogle Calendar
KPI CreatorAgent team · Ovy, Gary, Kewpie, Viktor, Jennifer
In Development, Refinements Needed
Claude Code
- Developed
- 20 Oct 2026Stage 4 approved 21 OctCoverage check 22 OctEnd to end · Agent team · Process
- Build time
- ~7h 50m3 days, 20 to 22 Oct
- Times used
- 43 test runs, plus the Adam Health live run (29 Oct)
- Total run time
- ~1h 41mMeasured on ClearCompare; Adam Health took ~1h 37m
- Version updates
- 43Commits to the kpi-creator repo, 20 to 28 Oct
What it does
- Turns a business problem into a finished KPI dashboard through 5 staged agents
- Gary interviews the stakeholder and picks the Top 5 KPIs with SMART targets
- Kewpie tests each KPI on the data; Viktor writes the story and chooses the charts
- Jennifer finishes and checks the dashboard; Ovy runs the stages and approval gates
Results
- ClearCompare test case through Stages 1 to 4, each one approved
- Coverage check: 15 of 22 stakeholder brief items addressed, 8 next steps proposed
- Animated replay of the run, built from recorded save and commit times
- Kestrel Medical closed 24 Oct; Lumenwire Audio closed 27 Oct with stakeholder sign-off
- Lumenwire Audio sealed check: 10 of 20 (Partial)
- First live run: Adam Health Base to Build dashboard on Apple Health data, signed off 29 Oct
Obstacles
- Kestrel run 2 stopped: stakeholder handbacks leaked answers; fixed with a one-word rule
- Kewpie stalled on long context; fresh sessions fixed it
- Agent run time not logged directly; measured from save and commit times
- Lumenwire run used 137.66M tokens against a 25M target
Skills and tools
Claude Code agentPythonHTML, CSS, JSSVG chartsMiro
UX Standards ReviewSkill · /review-ux-standards
Successfully Completed & Running
Claude Code
- Developed
- 19 Oct 2026
- Build time
- ~25mBuilt with HTML Standards Review
- Times used
- ~4AI Build Log ×2, portfolio ×1, 29 Oct ×1
- Total run time
- ~15mEstimate
- Version updates
- 119 Oct
What it does
- Checks a page against your UX Standards Reference at 5 screen sizes
- Counts touch targets under 44px and checks sideways scroll, readability and dark mode
- Scores the UXPin design principles
- Lists fixes as a before and after table; changes nothing until you confirm
Results
- AI Build Log: no sideways scroll at any size after fixes
- Portfolio fixes applied after review
Obstacles
- Renamed from /ux-standards-review on 19 Oct
- Built-in browser often reports the wrong width; checks now use fixed-width frames
Skills and tools
Claude Code skillBrowser pane
HTML Standards ReviewSkill · /review-html-standards
Successfully Completed & Running
Claude Code
- Developed
- 19 Oct 2026
- Build time
- ~25mBuilt with UX Standards Review
- Times used
- ~4Portfolio ×1, AI Build Log ×2, 29 Oct ×1
- Total run time
- ~20mEstimate
- Version updates
- 119 Oct
What it does
- Checks HTML structure, WCAG 2.2 AA accessibility, CSS, JavaScript, speed and security
- Measures contrast in light and dark mode and does a keyboard pass
- Cites the WHATWG spec section for each HTML finding
- Lists fixes as a before and after table; changes nothing until you confirm
Results
- Portfolio: landmarks, real buttons and contrast fixed (19 Oct)
- AI Build Log: focus rings, font weights and phone layout fixed
Obstacles
- Renamed from /html-standards-review on 19 Oct
- Flagged a private coach app link on a public page; now links the public demo
Skills and tools
Claude Code skillBrowser pane
Digital Pollen AnimationSkill · /add_animation_background
Successfully Completed & Running
Claude Code
- Developed
- 19 Oct 2026
- Build time
- ~30mFrom the AI Build Log header animation
- Times used
- 3
- Total run time
- ~2m
- Version updates
- 219 and 21 Oct
What it does
- Adds a slow "digital pollen" animation behind any section of a page
- Previews subtle, medium and strong before you choose
- Always adds one highlight grain in a colour outside the page palette
- Respects reduced-motion settings and never publishes
Results
- Used 3 times, including the AI Build Log and portfolio headers
Obstacles
- First highlight grain was hard to see on the portfolio; changed to blue
- Canvas grew on every resize; fixed with explicit sizes
Skills and tools
Claude Code skillHTML, CSS, JSCanvas animation
AI Build Log UpdaterSkill · /update_AI_log
In Development, Refinements Needed
Claude Code
- Developed
- 19 Oct 2026First run 22 Oct
- Build time
- ~25m
- Times used
- 322, 27 and 29 Oct
- Total run time
- ~3mScan and list, before approval
- Version updates
- 422, 26 and 27 Oct
What it does
- Scans skills, agents, scheduled tasks, cloud routines, repos and page views
- Compares the scan with this log and lists new builds and updates
- Changes nothing until you approve the numbered list
- Never touches GitHub until you say "Update Github"
Results
- First run 22 Oct: 6 new builds and 15 updates found and approved
- Portfolio sync step added 26 Oct
- Second run 27 Oct: 3 new builds and 14 updates found and approved
Obstacles
- GitHub sign-in is not available inside Python, so the repo list runs from the shell
- Page-view count failed until the site address format was fixed
Skills and tools
Claude Code skillPythonGitHub CLIArtifact database
AI Brain GitHub SyncPrivate repo · my-ai-brain
Successfully Completed & Running
Claude Code
- Developed
- 18 Oct 2026
- Build time
- ~30mNot measured
- Times used
- 79Commits pushed, 18 to 29 Oct
- Total run time
- SecondsPull at start, push at wrap
- Version updates
- 1Pull on resume and secret check added
What it does
- Keeps the whole AI Brain folder in a private GitHub repo
- Pulls the latest copy automatically when a session starts
- /wrap commits and pushes at the end of each session
- Checks for secrets before every commit
Results
- 71 commits since 18 Oct
- Cloud and phone sessions can use every skill and file
Obstacles
- PC-only jobs (speed test, WhatsApp, Chrome) cannot run in cloud sessions
- API keys stay in git-ignored files, with example files committed instead
Skills and tools
GitHubClaude Code hooks
Agent SimulationSkill · /create-kpi-simulation
Successfully Completed & Running
Claude Code
- Developed
- 23 Oct 2026Part of KPI Creator
- Build time
- ~1h 30mEstimate, 3 sessions, 23 to 27 Oct
- Times used
- 227 and 29 Oct
- Total run time
- ~14m
- Version updates
- 426, 27, 28 and 29 Oct
What it does
- Replays a KPI Creator run as an animation, with the five agents in a circle
- Shows each output as it was made, with handovers at each approval
- Timed from measured agent run time
- Saves to the project's deliverables folder and never publishes
Results
- Agent Simulation built for a KPI Creator test run on 27 Oct
Obstacles
- Agent run time is not logged directly, so it is measured from save and commit times
Skills and tools
Claude Code skillPythonHTML, CSS, JS
KPI Token ReportSkill · /kpi-token-report
Successfully Completed & Running
Claude Code
- Developed
- 24 Oct 2026Part of KPI Creator
- Build time
- ~45mEstimate, 24 to 26 Oct
- Times used
- 0By command; token pages made in KPI sessions
- Total run time
- None yet
- Version updates
- 0Unchanged since 26 Oct
What it does
- Measures token use for a KPI Creator run, split by agent
- Compares any or all runs side by side
- Builds a comparison page in the KPI Creator token reports folder
- Never publishes
Results
- Token comparison page live: ClearCompare, Kestrel Medical and Lumenwire Audio (10 updates, 24 to 27 Oct)
Obstacles
- Estimates had to be replaced with actuals once each run finished
Skills and tools
Claude Code skillPythonToken reports
KPI End-to-End UpdaterSkill · /update-kpi-end-to-end
Successfully Completed & Running
Claude Code
- Developed
- 27 Oct 2026Part of KPI Creator
- Build time
- ~30mEstimate, 1 session
- Times used
- 0
- Total run time
- None yet
- Version updates
- 0Built 27 Oct
What it does
- Adds a finished KPI Creator test case to the public end-to-end page
- Publishes the case's outputs alongside the earlier runs
- Keeps the sealed test bank off the public page
- Pushes to GitHub Pages only when you approve
Results
- Not used yet; built after the Lumenwire Audio run closed
Obstacles
- None recorded yet
Skills and tools
Claude Code skillPythonGitHub PagesGitHub CLI
Cardio StoryScrollytelling page · "Paddling to a fitter heart"
Not set
Claude Code
- Developed
- 28 Oct 2026Live 29 Oct
- Build time
- ~5h 25m2 sessions, 28 to 29 Oct
- Times used
- Not trackedNo visit counter yet
- Total run time
- Always onWeb page, no run time
- Version updates
- 9Commits to the public repo, 29 Oct
What it does
- Shows how VO2 Max rose with dragon boating, from Apple Health data
- Original and Heart Monitor styles on one page, with a toggle
- Uses Apple's official fitness levels, with a fitness dial badge
- Play scrolls through each chapter, then draws its chart
Results
- Live at adamaziz1979.github.io/paddlingtoafitterheart
- Average VO2 Max up 9.0% from 2025 to 2026
- Above average fitness in 2 of 8 paddling months, first in Sep 2026
Obstacles
- Festival style tried and dropped; two styles kept
- Apple does not publish women's ranges; estimated from FRIEND and labelled
- Health export kept off GitHub; the build script reads it locally
Skills and tools
Claude CodePythonHTML, CSS, JSSVG chartsApple Health exportGitHub Pages
Overall assessment
- Biggest builds: the portfolio website (~11h 15m), the Dragonboat Seat Planner (~8h 30m) and the KPI Creator (~7h 50m).
- The portfolio feeds evidence into every Build Company Brief.
- The Seat Planner went from prototype to v70 in 3 days.
- The KPI Creator ran its first test case through 4 of 5 stages in 3 days.
- Strongest result: a full interview-prep chain that has produced 5 briefs so far.
- The ChatGPT research skill moved to Claude Code on 28 Sep.
- Claude Code added an HTML report and publishing.
- Job scanning runs in 3 places: all three show refreshed ads as new.
- Daily Job Scanner in Claude Code: daily schedule off since 14 Oct.
- ChatGPT 6am and 5pm tasks: still producing output.
- Settling on one daily scan would remove repeated recommendations.
- Platform split: Claude Code handles files, Gmail drafts, WhatsApp, GitHub and PC schedules.
- ChatGPT handles the recurring job-search briefings.
- Gaps: no recorded ChatGPT build effort and no measured time saved.
- No link yet from tools to interview outcomes.
- The 5pm task's schedule status is not confirmed.
How the numbers were worked out
- Developed: the date the skill or agent file was first written, checked against Claude session logs.
- Build time: an estimate of hands-on time in each session.
- Counts active minutes from your first request to the last change.
- Breaks longer than 5 minutes are capped at 5 minutes.
- Not counted: time spent using a build after it was finished.
- For example, publishing reports or doing job scans.
- Supporting pieces built separately, such as the speed-test collector (about 45 min, 19 to 25 Aug).
- Shared sessions: some builds were made together, so their time is split.
- Company Research Agent, Build Company Brief and Build Company Brief HTML Report: 28 and 29 Sep, about 1h 50m.
- New GitHub Site and Update Live Site: about 20 min together.
- Portfolio website: time covers the personal brand sessions, 26 Jul to 20 Oct.
- Includes small related pieces: a role-themes diagram and a Data Strategy & Evolution link.
- Treat it as within plus or minus 30%.
- Dragonboat Seat Planner: time covers the Claude Code sessions, 12 to 19 Oct.
- Versions 1 to 17 were built earlier in a claude.ai chat; that time is not counted.
- Use is not tracked: the app runs on each coach's phone with no analytics.
- KPI Creator: time covers the KPI Creator sessions, 20 to 22 Oct.
- Refinements 23 to 27 Oct are not counted.
- Run time is measured agent time on the ClearCompare test case.
- Builds added 22 Oct: found by the AI Build Log Updater.
- The review skills' uses include runs under their old names.
- AI Brain GitHub Sync build time was not measured.
- Build added 29 Oct: Cardio Story, found by the AI Build Log Updater.
- Build time counts its 2 sessions, 28 and 29 Oct.
- The Adam Health KPI run counts as a KPI Creator use, not a new build.
- Builds added 27 Oct: three KPI Creator skills, found by the AI Build Log Updater.
- Their build times are estimates from editing sessions.
- AI Build Log time from 25 to 26 Oct is also an estimate.
- ChatGPT builds: dates, results and obstacles come from your ChatGPT inventory (18 Oct).
- Build dates are when the related chat was created.
- Build time was not recorded.
- Results and obstacles: counts come from output folders, run logs, trackers and GitHub, as at 29 Oct.
- Obstacles are problems raised in build sessions, and how each was handled.
- Status: set by Adam, as at 29 Oct 2026.
- Times used and run time: counted from session logs and scheduled and cloud run history, to 29 Oct.
- A use is one slash command or scheduled run.
- Run time is Claude's working time, not time spent waiting on you.
- Portfolio views come from GoatCounter.
- ChatGPT counts assume one use a day and have not been checked.
- Version updates: how many times each build changed after it was first built.
- Counted as separate editing sessions in the session logs.
- The portfolio and seat planner use their published version history.
- ChatGPT figures come from your inventory.
- Precision: figures are rounded to the nearest 5 minutes. Treat them as within plus or minus 25%.