Unstuck Coach

A folder-based whole-person executive-function accessibility coach for people who need help starting, switching, remembering, regulating, capturing, recovering, and closing loops without shame.

Drop this folder into a Claude Project, Codex workspace, Antigravity-style AI IDE project, or local-model context and ask the coach to help with the life loop in front of you: work, home, body, admin, inbox, calendar, communication, transition, re-entry, or shutdown.

The core idea: Unstuck Coach acts as portable executive-function accessibility. It externalizes state, context, next action, capture, time, transition, and closure so the person does not have to carry all of that internally.

Who This Is For

Unstuck Coach is for people who already have values, obligations, care, and intelligence, but lose traction at the executive-function layer:

  • Too many tasks or life loops in working memory.
  • Difficulty starting a known task, chore, message, errand, calendar check, inbox pass, or repair.
  • Time estimates that collapse on contact with reality.
  • Hyperfocus followed by crash.
  • Shame after avoidance.
  • Message, conflict, or communication-threat spirals.
  • Inbox piles, reply debt, calendar drift, and scheduling friction.
  • Unclear re-entry after an interrupted session.
  • Raw ideas that need capture without turning into a new planning spiral.
  • Body-state needs that must be handled before planning can work.
  • End-of-day or transition points that need clean closure.

What This Coach Does

It helps the user make the next move without turning executive dysfunction into a character flaw.

It does not just answer questions about ADHD, productivity, or motivation. It coaches in the moment:

  • Checks state before planning.
  • Reduces options to one next action.
  • Holds the rest of the list.
  • Offers body-doubling scripts.
  • Creates re-entry breadcrumbs.
  • Uses shame-free return protocols.
  • Captures raw ideas without turning them into a lecture.
  • Sorts raw brain dumps into one next move plus parked context.
  • Uses a bounded dopamine menu when the user needs activation fuel before starting.
  • Makes time visible with buffers, hard stops, and start anchors.
  • Uses a tangent firewall: chase, bookmark, or discard.
  • Switches strategy after repeated failed attempts.
  • Resets messy systems without shame.
  • Turns calendar and inbox management into small visible passes instead of a demand to process everything.

Try It Quickly

  1. Open the canonical landing page: https://unstuck.kyanitelabs.tech/.
  2. Try the live GLM chat demo: https://unstuck.kyanitelabs.tech/chat/.
  3. The live demo is backed by GLM-5.1 through Z.AI's international coding-plan endpoint with reasoning enabled and a 1200-token completion budget.
  4. Open START_HERE.md.
  5. Load the folder into a Claude Project or another AI workspace.
  6. Paste PROJECT_INSTRUCTIONS.md.
  7. Try: I need a coach to get started on this.
  8. Score the first response with FIRST_REPLY_SCORECARD.md.

The landing page gives the visual version of this path. The receipts are available when you want to inspect source files, transcripts, or evals.

Receipt Room: Folder Map

This section is for inspection after the human value is clear.

unstuck-coach/
├── robots.txt
├── sitemap.xml
├── llms.txt
├── LICENSE
├── AGENTS.md
├── START_HERE.md
├── PRODUCT_THESIS.md
├── FIRST_RUN.md
├── FIRST_REPLY_SCORECARD.md
├── COMPETITION_RULES_TRACE.md
├── ICM_TRACE.md
├── WRITEUP.md
├── HANDOFF_CARD.md
├── JUDGE_SCORECARD.md
├── JUDGE_FAQ.md
├── JUDGE_BRIEF.md
├── PROJECT_INSTRUCTIONS.md
├── PUBLICATION_CHECKLIST.md
├── WALKTHROUGH.md
├── PITCH_REEL.md
├── RECEIPTS.md
├── SUBMISSION.md
├── identity.md
├── rules.md
├── examples.md
├── docs/
│   └── judge-walkthrough.md
├── demo/
│   ├── before-after.md
│   ├── transcript-pack.md
│   └── whole-person-tour.md
├── landing/
│   ├── assets/
│   │   ├── unstuck-coach-logo.png
│   │   ├── unstuck-handoff-card.svg
│   │   └── unstuck-admin-bridge.jpg
│   ├── index.html
│   ├── evidence.html
│   ├── reel.html
│   ├── styles.css
│   ├── reel.css
│   └── app.js
├── scripts/
│   ├── prepare-publication-link.mjs
│   ├── public-bundle-files.mjs
│   ├── build-evidence-page.mjs
│   ├── render-review-screenshots.mjs
│   ├── verify-landing-accessibility.mjs
│   ├── verify-source-notes.mjs
│   ├── verify-submission-copy.mjs
│   ├── verify-submission-surfaces.mjs
│   ├── verify-pitch-reel.mjs
│   ├── verify-reel-page.mjs
│   ├── verify-judge-faq.mjs
│   ├── verify-judge-scorecard.mjs
│   ├── verify-judge-brief.mjs
│   ├── verify-competition-rules-trace.mjs
│   ├── verify-product-thesis.mjs
│   ├── verify-icm-trace.mjs
│   ├── verify-first-run.mjs
│   ├── verify-first-reply-scorecard.mjs
│   ├── verify-start-here.mjs
│   ├── verify-landing-copy.mjs
│   ├── verify-transcript-pack.mjs
│   ├── verify-first-reply-acceptance.mjs
│   ├── verify-whole-person-tour.mjs
│   ├── verify-mode-router.mjs
│   ├── verify-console-behavior.mjs
│   ├── verify-eval-coverage.mjs
│   ├── verify-admin-ops-playbooks.mjs
│   ├── judge-quick-proof.mjs
│   ├── verify-public-bundle.mjs
│   ├── verify-final-privacy-scan.mjs
│   ├── verify-publication-ready.mjs
│   ├── verify-github-public-url.mjs
│   ├── final-review-smoke.mjs
│   ├── verify-clean-public-stage.mjs
│   ├── build-public-bundle.mjs
│   └── stage-public-repo.mjs
├── evals/
│   ├── red-face-tests.md
│   └── research-to-behavior-checklist.md
├── reference/
│   ├── coaching-protocols.md
│   ├── admin-ops-playbooks.md
│   ├── mode-router.md
│   ├── signal-map.md
│   ├── safety-boundaries.md
│   └── source-notes.md
└── README.md

How To Use

Search And AI Discovery

When the repository is made public, set the repository About homepage to the Kyanite landing page:

https://unstuck.kyanitelabs.tech/

The front-facing discovery files are:

  • robots.txt for crawler access and sitemap discovery.
  • sitemap.xml for the landing page, live chat demo, evidence reader, and pitch reel URLs.
  • llms.txt for answer engines, AI search, and LLM citation context.
  • landing/index.html for browser metadata, Open Graph/Twitter previews, canonical URL, and structured data.

Claude Project

  1. Create a Claude Project.
  2. Add this whole folder as project knowledge.
  3. Paste PROJECT_INSTRUCTIONS.md into the Claude Project instructions.
  4. Start a chat with:
You are Unstuck Coach. Read identity.md, rules.md, examples.md, and reference/. Coach me through the life loop in front of me. If my first message is vague, ask one state-calibrating question. If I name a stuck signal, route it directly.

Codex

Open the folder as the Codex workspace. AGENTS.md is the native coding-agent entrypoint, so Codex should load the coach contract from the folder before it answers coaching prompts.

codex -C unstuck-coach "I need a coach to get started on this."

If your coding agent does not automatically read AGENTS.md, paste the same start prompt from the Claude Project path first.

Antigravity Or Another AI IDE

Create a project from the Unstuck folder. Keep file access scoped to this project, then ask the workspace to read PROJECT_INSTRUCTIONS.md, START_HERE.md, identity.md, rules.md, examples.md, and reference/ before coaching.

Use the same first chat message from the Claude Project path.

Local Models

Use a local runner with enough context to hold the operating files. If it cannot ingest the whole folder, paste PROJECT_INSTRUCTIONS.md first, then add:

  • identity.md
  • rules.md
  • examples.md
  • reference/coaching-protocols.md
  • reference/signal-map.md
  • reference/safety-boundaries.md

Then start with one cold prompt from START_HERE.md.

Quick Evaluation Path

This is a shortcut, not the product boundary. Unstuck's scope remains whole-person executive-function accessibility across life loops; this path only makes the first reply easy to test.

Open START_HERE.md for the shortest path, JUDGE_BRIEF.md for the one-page above-the-brief case, then landing/index.html for the visual walkthrough.

Open FIRST_RUN.md if you want the exact cold-start receipt before reading the whole folder.

Use the live demo button on the landing page to test a real GLM-backed stuck point. The protocol verifier remains in scripts/verify-console-behavior.mjs as proof of the routing logic.

Open docs/judge-walkthrough.md for a longer evaluation path.

The quick test:

  1. Load the folder into a Claude Project.
  2. Try: I need a coach to get started on this.
  3. Try: I need to pay the bill, eat something, and answer the text, but I am frozen.
  4. Try: My inbox and calendar are a mess and I do not know what is real.
  5. Try: idea: make a shutdown checklist for Sunday nights
  6. Try: I tried the same plan three times and failed every time.

If Unstuck gives a productivity article, it failed. If it gives one state-aware next move and closes the loop, it is behaving like a coach.

Best First Prompts

I cannot start this task.

I need a coach to get started on this.

I have too many things in my head. Help me sort them.

My inbox and calendar are a mess and I do not know what is real.

I am coming back after avoiding this project.

That message makes me feel like I did something wrong.

Body double me through the first concrete move.

idea: make a shutdown checklist for Sunday nights

I tried the same plan three times and it keeps failing.

I have 25 minutes before I leave and think I can finish the whole pile.

Proof Artifacts

  • landing/index.html gives the visual product front door: whole-person hero, coaching routes, Claude Project launch kit, Codex/AI IDE/local-model setup paths, live coaching demo bridge, folder-method map, and readable source-proof links.
  • robots.txt, sitemap.xml, and llms.txt give the public repo a crawler, answer-engine, and generative-engine discovery layer on the Kyanite domain.
  • landing/assets/unstuck-coach-logo.png gives the public brand mark used in the navigation and favicon.
  • landing/assets/unstuck-admin-bridge.jpg gives the admin-operations band a bitmap operating-map asset using the simplified PuenteWorks bridge/Mola visual language.
  • landing/reel.html gives a record-ready six-scene pitch reel page for the optional video/GIF proof layer.
  • AGENTS.md gives Codex, Claude Code, and AI IDEs a native coding-agent entrypoint for loading the coach contract from the folder.
  • START_HERE.md gives the 60-second route, exact start prompt, first-reply acceptance test, and fast cold prompts.
  • PRODUCT_THESIS.md explains why the folder is the product, why first contact is the public cold test, how whole-person EF accessibility maps to behavior, and where the coach stops.
  • ICM_TRACE.md maps the coaching system to staged, inspectable, editable, and auditable workflow architecture.
  • FIRST_RUN.md shows the exact cold-start prompt, expected first reply, tiny proof loop, first-message routing, and immediate fail patterns.
  • FIRST_REPLY_SCORECARD.md gives the fastest pass/fail gate for whether the first response actually coaches.
  • COMPETITION_RULES_TRACE.md maps every Week 5 rule and judging question to concrete evidence in the folder.
  • WRITEUP.md compresses the submission thesis into a short judge-readable argument.
  • HANDOFF_CARD.md gives a one-page operator handoff with use cases, good/bad behavior, gaps, and the cold-start test.
  • JUDGE_SCORECARD.md gives an 18-point scoring rubric with evidence links.
  • JUDGE_FAQ.md gives the shortest answers to likely Week 5 judging objections before a judge opens every file.
  • JUDGE_BRIEF.md gives a one-page above-the-brief argument so the winning case is not scattered across the folder.
  • PROJECT_INSTRUCTIONS.md gives paste-ready Claude Project instructions and a first-reply acceptance test so a stranger can load the coach without translating the README.
  • PUBLICATION_CHECKLIST.md gives the final publication gate without exposing local review notes.
  • WALKTHROUGH.md provides the longer recording/read-through script and shot list.
  • PITCH_REEL.md compresses the presentation layer into a verified 75-second judge reel.
  • RECEIPTS.md maps each public claim to the exact file that proves it.
  • demo/before-after.md shows how Unstuck Coach differs from generic productivity advice.
  • demo/transcript-pack.md gives cold-test transcript examples for the moments a judge is likely to try.
  • demo/whole-person-tour.md gives a six-stop cold tour across the full life surface.
  • evals/red-face-tests.md stress-tests shame, overload, brain dumps, dopamine-menu drift, inbox/calendar reality, time blindness, capture, tangents, hyperfocus, and safety boundaries.
  • evals/research-to-behavior-checklist.md maps the research foundation to actual coach behavior, including live-obligation rescue before cleanup.
  • reference/mode-router.md preserves the original multi-mode assistant insight as a public-safe stance portfolio: ally support, strategy, execution, memory keeping, and recovery closure.
  • reference/signal-map.md gives the whole-person operating surface map: food/body and activation fuel, calendar/inbox, messages/shame, home/admin loops, capture/re-entry, and closure/recovery.
  • reference/admin-ops-playbooks.md gives operational calendar and inbox playbooks for hard-anchor recovery, live-obligation rescue, reply debt, missed obligations, and scheduling friction without claiming account access.
  • scripts/prepare-publication-link.mjs dry-runs or writes the approved repository URL into SUBMISSION.md.
  • scripts/build-evidence-page.mjs renders the actual Markdown receipts into landing/evidence.html so proof clicks show readable source-backed content instead of hand-typed claim cards.
  • scripts/render-review-screenshots.mjs refreshes the landing, narrow/mobile first-glance, evidence reader, admin-band, first-run receipt, scorecard, FAQ, and reel screenshots for design approval using standard Playwright.
  • scripts/verify-landing-accessibility.mjs checks landing semantics, labeled controls, hash targets, reduced-motion handling, focus-visible treatment, and accessibility behavior wiring.
  • scripts/verify-source-notes.mjs checks the public source notes for competition fit, design lineage, research translation, portability, and private-provenance safety.
  • scripts/verify-submission-copy.mjs checks that the Skool comment draft stays within the required 2-3 sentence shape.
  • scripts/verify-submission-surfaces.mjs checks that the Skool draft and SUBMISSION landing version stay synchronized while the landing avoids a separate submission panel.
  • scripts/verify-pitch-reel.mjs checks that the 75-second pitch reel has six timed shots, a short voiceover, and no public-unsafe private/local references.
  • scripts/verify-reel-page.mjs checks that the record-ready reel page has six scenes, local refs, responsive CSS hooks, and no public-unsafe private/local references.
  • scripts/verify-judge-faq.mjs checks that the judge FAQ keeps eight short answers, evidence references, and no public-unsafe private/local references.
  • scripts/verify-judge-scorecard.mjs checks that the judge scorecard keeps nine criteria, a sequential fast path, evidence links, and no public-unsafe private/local references.
  • scripts/verify-judge-brief.mjs checks that the judge brief keeps the whole-person wedge, above-the-brief case, fast test, failure modes, ICM fit, evidence map, blocked state, and no public-unsafe private/local references.
  • scripts/verify-competition-rules-trace.mjs checks that the Week 5 rules trace keeps the required brief rows, judging-question rows, above-the-brief proof list, blockers, and no public-unsafe private/local references.
  • scripts/verify-product-thesis.mjs checks that the product thesis stays explicit about folder-first design, first-contact coaching, proof, and boundaries.
  • scripts/verify-icm-trace.mjs checks that the ICM trace stays public, evidence-backed, and tied to the proof layer.
  • scripts/verify-first-run.mjs checks that the first-run receipt keeps the cold prompt, expected first reply, proof loop, and fail patterns intact.
  • scripts/verify-first-reply-scorecard.mjs checks that the first-reply scorecard keeps its pass/fail criteria, fail patterns, and evidence links.
  • scripts/verify-start-here.mjs checks that the fastest judge path stays paste-ready.
  • scripts/verify-landing-copy.mjs checks that the landing launch-kit prompts, including the inbox/calendar recovery prompt, stay copy-ready.
  • scripts/verify-transcript-pack.mjs checks that the cold-test transcript pack stays complete.
  • scripts/verify-first-reply-acceptance.mjs checks that cold-test first replies name friction, give one move, ask for proof, and avoid generic advice patterns.
  • scripts/verify-whole-person-tour.mjs checks that the whole-person tour keeps six surfaces, pasteable prompts, proof checks, and immediate-fail signals.
  • scripts/verify-mode-router.mjs checks that the coach keeps five stance modes and rejects coding-only execution framing.
  • scripts/verify-console-behavior.mjs checks the proof-layer protocol classifications.
  • scripts/verify-eval-coverage.mjs checks red-face coverage and the research-to-behavior map.
  • scripts/verify-admin-ops-playbooks.mjs checks the calendar/inbox admin operations playbooks.
  • scripts/judge-quick-proof.mjs gives a publication-independent proof summary for the cold judge path, including original-Liam inbox/calendar behavior.
  • scripts/verify-public-bundle.mjs checks required files, landing proof/launch-kit text, local landing refs, public-safe text, submission copy, transcript completeness, and protocol behavior.
  • scripts/verify-final-privacy-scan.mjs gives the final post-link privacy pass a named command instead of leaving it as a manual grep ritual.
  • scripts/verify-publication-ready.mjs is the final posting gate after the public GitHub link is inserted; it rejects the old Week 3 repo and reruns the proof-layer checks, including source-note lineage, landing accessibility, whole-person tour coverage, and mode routing.
  • scripts/verify-github-public-url.mjs checks that the approved repository URL is visible through unauthenticated GitHub API access.
  • scripts/final-review-smoke.mjs runs the proof checks in one concise command, rebuilds the public payload unless skipped, and verifies whether the publication gate is expected to be blocked or ready.
  • scripts/verify-clean-public-stage.mjs stages the payload into a temporary separate folder, verifies it there, and removes the temporary target.
  • scripts/build-public-bundle.mjs exports the clean public repo payload after the source bundle verifies.
  • scripts/stage-public-repo.mjs dry-runs or writes the verified payload into a clean public repository folder.

Verify The Public Bundle

Run:

node scripts/judge-quick-proof.mjs
node scripts/verify-source-notes.mjs
node scripts/verify-landing-accessibility.mjs
node scripts/verify-mode-router.mjs
node scripts/verify-admin-ops-playbooks.mjs
node scripts/verify-whole-person-tour.mjs
node scripts/verify-judge-brief.mjs
node scripts/verify-public-bundle.mjs
node scripts/verify-final-privacy-scan.mjs

The quick proof reports the cold-start path, first-reply gate, transcript evidence, whole-person tour, live-demo bridge, protocol behavior verifier, stress evals, admin operations playbooks, source-notes lineage proof, research-to-behavior proof, product thesis, ICM trace, judge FAQ, scorecard, concise judge brief, and fastest cold prompts without requiring the final public GitHub link. The bundle verifier checks required files, local landing-page links/assets, product thesis, source notes, Week 5 rules trace, ICM trace, pitch reel, record-ready reel page, judge FAQ, judge scorecard, judge brief, first-run receipt, first-reply scorecard, start-here prompt readiness, landing accessibility, landing copy controls, mode routing, Skool comment shape, synchronized submission surfaces, transcript completeness, whole-person tour coverage, first-reply acceptance, protocol behavior, public-safe checklist text, emoji/symbol-range leakage, private provenance patterns, and the standalone final privacy scan. Before final publication it may warn that the GitHub link is still pending; after the public repo link is inserted, those warnings should be gone.

To prove the payload can stage into a separate clean repository folder, run:

node scripts/verify-clean-public-stage.mjs

Build The Public Repo Payload

Run:

node scripts/build-public-bundle.mjs

This creates an ignored export at output/public-bundle/unstuck-coach/. After final link approval, copy that exported folder into the clean Week 5 public repository and run the verifier again from inside the copied repo.

Dry-run staging into a separate clean repository folder:

node scripts/stage-public-repo.mjs --target ../unstuck-coach-week5-public

After reviewing the target, write it:

node scripts/stage-public-repo.mjs --target ../unstuck-coach-week5-public --write

Final Publication Gate

After final approval and once the public GitHub repository exists, dry-run the link insertion:

node scripts/prepare-publication-link.mjs --url https://github.com/OWNER/REPO

After reviewing the dry-run output, write it:

node scripts/prepare-publication-link.mjs --url https://github.com/OWNER/REPO --write

Then run:

node scripts/verify-publication-ready.mjs
node scripts/verify-github-public-url.mjs
node scripts/verify-final-privacy-scan.mjs

Both checks should report status: "ready" after the approved repository URL is inserted. Before that handoff, this gate is expected to report status: "blocked". It should also stay blocked if the link points at the old Week 3 repository or at any repository unauthenticated GitHub API access cannot see.

For one-command final review before the link exists, run:

node scripts/final-review-smoke.mjs --expect-blocked

After the final public link is inserted, run:

node scripts/final-review-smoke.mjs --expect-ready --skip-build

The Bar

If the coach gives you a long productivity article, it failed.

If it asks one useful question, names the friction without shame, and helps you take one concrete next step, it is doing its job.

Safety Note

This is a coaching scaffold, not medical care or therapy. It does not diagnose, treat, recommend medication, or replace professional support. If a user is in crisis or cannot stay safe, follow reference/safety-boundaries.md.