yingaowang-casia/shushu-projectproof

ProjectProof for Codex is a Codex skill, meaning a small extension that plugs into the Codex coding assistant. The README is written in Chinese and aims it at students or junior engineers who did small internship projects in areas like AI agents, RAG, algorithms, data cleaning, or frontend and backend, and who now want to interview at large tech companies. The author calls it a "project evidence officer" and is careful to say that it does not invent experience. It only reshapes real work so that it can survive deeper questioning in an interview. The install path is a curl command that pulls install.sh from GitHub and pipes it to bash. The script copies a folder called bigtech-interview-project-upgrader into ~/.codex/skills, and after restarting Codex the user types "use ProjectProof for Codex" along with the target job description, their current resume text, and notes about what they actually did, what state the project is in, what evidence they have such as code, logs, screenshots, metrics or bad cases, and what questions they are afraid of. The skill then runs a fixed sequence. First it sorts each piece of project work into categories: actually built, deeply involved, only aware of, can be written after extra work, and currently not writable. Next it asks for evidence behind every strong claim and ranks each claim from C0 to C3 based on how solid that evidence is. Then it looks for the engineering gap between a toy demo and a production system, listing missing pieces such as tool schemas, eval sets, monitoring, retries, or rollback. After that the skill builds a question tree that mimics how a senior interviewer would keep drilling, and for each question it drafts three answers: a risky one, a safer one, and a strong one. It closes with an SP and SSP style scorecard across job match, technical depth, contribution clarity, evidence strength, engineering closed loop, and stability under followups. The outputs include three versions of resume bullets (conservative, standard, and strong), a followup tree, a two to three minute project pitch, and a checklist of evidence the candidate should gather in half a day, one day, three days, or one week before applying.