← ClaudeAtlas

project-intellisted

Use when starting, resuming, or advising on any project — settled results from real project history: proven approaches, known dead ends, and era-tagged evidence, so nothing already decided gets re-searched.
SenjuWoo/Ultimate-AI-Starter-Bundle · ★ 7 · AI & Automation · score 78
Install: claude install-skill SenjuWoo/Ultimate-AI-Starter-Bundle
# Project Intel The operator's project history already proves a large number of approaches and disproves others. Re-searching settled questions burns tokens and re-introduces known-bad paths. This skill carries the settled results, one card per project class. For a creative project whose look, tone or topology must survive a restart, pair this card with `creative-direction`: it owns the Vision Contract and feature ledger, and this skill keeps the settled evidence those decisions rest on. ## Use 1. Identify the project class and load the **one** card that matches the router below. Never load all cards. 2. Treat every card line as evidence with a date, not eternal truth. Cards state what was observed, when, and how strongly. 3. **Freshness rule.** Judge evidence by its date, version, reproducibility and relevance, not the model that wrote it. Old tool output can still prove a stable invariant; a new model's assertion is not runtime proof. Model, price and tool facts rot fastest: verify live before advising on them. 4. When new work proves or disproves something durable, append a dated line to the matching card. One line, with the date and the strength of evidence. **Write a workflow-scoped lesson, never a bare verdict:** workflow/build -> what was tried -> result -> cause -> what worked instead -> evidence (strength, scope) -> recheck when. "X does not work" is not a lesson; "in workflow W, X failed because C, and Y worked" is. A failure seen i