searchlisted
Install: claude install-skill varuntej07/tej-stack
# Search
Investigate: the user's question.
Strictly read-only: no edits to any tracked file, no builds, no git writes. The only file you
write is the report. Read `.context/tej-core.json` for `repos` and never read a `never_touch`
folder.
1. Use the `CLAUDE.md` lookup table to pick the `.context/1-docs/ARCH-*.md` docs; read them first
so the search starts from the documented design, then verify every claim against live code.
2. Search every repo in `repos`. Follow the call across every boundary (IPC, REST, websockets,
queues, databases, object storage).
3. Write `.context/4-scratchpad/SEARCH-<slug>.md` with:
- **Answer first**: the direct answer or number in the first lines.
- **The story**: a short first-person paragraph of how the path works, in plain words.
- **ASCII block diagram** of the path found.
- **Call graph**: ordered `file:line function` hops, one per line, with the transport at each
boundary.
- **Latency and storage audit**: every timeout, deadline, cache and database or storage read
and write on the path, with file:line; "unmeasured" where no number exists.
- **Resilience gaps**: the path's profile (FastFail or Resilient, from
`resilience-profiles.md`) and every deviation (retry on a FastFail path, missing timeout,
missing fallback, swallowed cancellation, cause-dropping catch).
- **Doc drift**: anything the ARCH doc says that the code contradicts.
4. Reply with the answer, the report path and the top three find