GitHub first

Search before building. Someone has usually built 80 percent of it, and a repo with a thousand stars has had its edge cases found by other people.

Steps

Name the job in 3 to 6 search terms. The job, not the tech (“pdf invoice parse”, not “python library”).
Search. Run two or three variations:
```
gh search repos "<terms>" --sort stars --limit 10 \
  --json fullName,description,stargazersCount,pushedAt,license
```
Also check local first: ~/.claude/skills, ~/.claude/plugins, ~/Repos, and the memory index. A thing he already has beats any repo.
Filter. Keep repos with 1,000+ stars (or 200+ for a niche job), a push in the last 6 months, and a licence that allows reuse (MIT, Apache, BSD). Drop anything with no licence.
Read the top 3. README plus a look at the main source files (gh api repos/OWNER/REPO/readme). Security skim: install scripts that fetch remote code, curl | bash, obfuscation, credential handling.
Decide, one of three:
- Adopt: install it as is. Only if it fits with little glue.
- Adapt: copy the approach or the relevant files, keeping the licence notice, and build the rest.
- Build: nothing fits. Say why in one line per repo checked.
Report: a table of repos checked (name, stars, last push, verdict) and the decision. Put it in the plan or PRD if one is being written.

This is a 5 to 10 minute step, not a research project. Cap at 3 searches and 3 READMEs unless the user asks for more.
Never install anything in this step. Adopting is a separate, approved action.
Stars can be bought. A repo with many stars but few forks, no issues and one commit is suspect.

gh search returns nothing: the terms are too specific. Drop one term.
Only old repos: fine to adapt the idea, not to install unmaintained code.
gh auth error or 404 on a repo he owns: another session may have switched the active gh account (gh auth status).