G-WARP — SINGLE ENTRY-POINT TEAM WORKFLOW
Owner: Creapter1
Updated: 2026-10-18
Purpose: one clear entry point. This file routes work; it does not authorize live integration.

START HERE — READ ONLY WHAT YOUR TASK NEEDS
1. ../AI-TEAM.txt — team roles, hard rules, chronological work log. This file was restored from known-good commit 8f5d9a4ca69db8b1f314cec683e628883177292c after an accidental truncation.
2. ../TODO.txt — product priorities and requirements.
3. CONTINUOUS-WORK-QUEUE.txt — choose one unclaimed bounded task.
4. ../ENGINE-MAP.txt — before proposing code, then verify the exact hooks in current source.
5. ../MERGE-QUEUE.txt — candidate status only; “PICK” is not proof of safety or approval.
6. Read the selected task's own report/notes only. Do not reread every audit or the whole app.

AUTHORITY / SAFETY
- Creapter1 is the owner and final approver.
- Claude may integrate only when the active Claude model label is exactly “Claude Opus 5.5” and the owner has authorized the task. If the label is missing/unclear/different, stop integration and do read-only work.
- Never rewrite app.html or index.html wholesale. Do not modify either file unless the owner explicitly authorizes the exact change. Prefer small isolated candidate files in the author AI's folder.
- Do not edit another AI's folder. Keep candidates disconnected from live behavior until authorized integration.
- Never delete a feature, claim a test that did not run, or call syntax checks behavioral proof.
- Preserve uncertain/older work; mark it SUPERSEDED or ARCHIVE rather than deleting it unless the owner specifically approves deletion.
- Record baseline commit + file hashes before any authorized source edit. Log exact tests, results, risks and rollback.

TASK LOOP
Claim one task -> state issue and measurable acceptance test -> inspect only relevant source -> reproduce -> make one minimal candidate/fix if justified -> run cheap checks then targeted runtime tests -> record evidence -> update task/merge status -> checkpoint -> continue to the next unclaimed in-scope task.
If blocked, state the exact blocker and do a useful independent task; do not invent results or pad time.

EVIDENCE LABELS
SOURCE-VERIFIED = exact current source inspected.
STATIC-ANALYSIS = reasoning/review only; not executed.
REPORTED TEST = another person/tool reports a test; not independently reproduced.
BROWSER-VERIFIED = test personally run in browser with steps/results recorded.
NOT TESTED = no supporting execution evidence.

ACTIVE WORKFLOW DOCUMENTS (single purpose each)
- This file: entry point and routing.
- ../AI-TEAM.txt: roles, rules, and chronological team log.
- ../TODO.txt: product requirements/priorities.
- CONTINUOUS-WORK-QUEUE.txt: next bounded tasks.
- ../ENGINE-MAP.txt: source/API map; verify against current source.
- ../MERGE-QUEUE.txt: candidate disposition.
- CLAUDE-EXECUTION-PLAN-2026-10-18.txt: current Claude long-run/low-token procedure. Do not use older Claude plans as active instructions.
- CLAUDE-CHECKPOINT.txt: preliminary metadata checkpoint exists; Claude must reverify it against the actual working copy and overwrite it with real test evidence. Never guess hashes or tests.
- AI-STATUS.txt: the one current team status and Claude handoff.
- AI-DELIVERABLE-INVENTORY.txt and AI-PROGRESS-PREVIEW-2026-10-18.txt: archived summary pointers; use their linked detailed reports only when a specific gap requires them.
- FUTURE-IDEAS.txt: unapproved ideas; preserve, do not implement.
- Each AI's own folder/AI-CHAT-LOG.txt: append only to that AI's own log. Gemini's unified workspace includes visuals, the browser-extension prototype, and inherited research/A-B/C/test-harness work under gemini/lab/. The DeepSeek role is retired; use team/TODO-GEMINI.txt and do not create new deepseek/ files.

DUPLICATE / STALE DOCUMENT POLICY
- The current Claude execution plan is the 2026-10-18 version 2.1. team/CLAUDE-PLAN-2026-10-17.txt is historical and must not override it.
- Older reviews, inventories and reports are historical evidence, not additional rules. Update the single-purpose active file instead of creating another broad plan/audit.
- Keep one current status per task in the queue/merge queue. Cross-link reports rather than copying the same analysis into multiple files.
- Do not mass-delete files during this cleanup. First identify exact duplicates, preserve unique findings, update links, then archive/delete only after checking references and owner constraints.
- Recent Buffy add-ons/settings ZIP: authorship is not proven by archive contents. Review is in chatgpt/BUFFY-ADDONS-SETTINGS-REVIEW-2026-10-18.txt; no browser test or integration was performed. Do not label it Gemini/DeepSeek-authored without evidence.

FIRST TASK FOR CLAUDE
Verify active model label and owner authorization. Read this file, AI-STATUS.txt, AI-TEAM.txt's rules and relevant recent log entries, TODO.txt, CONTINUOUS-WORK-QUEUE.txt, ENGINE-MAP.txt, MERGE-QUEUE.txt, and CLAUDE-CHECKPOINT.txt. Reconfirm the actual working-copy app.html/index.html baseline and hashes. Then follow the current 2026-10-18 Claude plan; prioritize reproducing the reported freeze/slow Stop issue. The reported 81,000-clone/45-second case is not yet a reproduced result. Do not touch app.html/index.html without the owner's explicit authorization.