jmb-research
Spin up a background agent to do the research, so you keep working while it reads. Start it with a fresh context window.
Its job:
-
Investigate the question against primary sources (official docs, source code, specs, first-party APIs), not a secondary write-up of them. Follow every claim back to the source that owns it.
-
Write the research report: a single Markdown file citing each claim's source, in the language of the "Documentation" scope of the Languages table in AGENTS.md. Name it
<YYYY-MM-DD>-<slug>.md, with today's date. When it cites numbers that helper scripts produced, make it a folder<YYYY-MM-DD>-<slug>/instead, with the report asREADME.mdand the scripts beside it, written in the language of code. -
Open it with its snapshot line, the first line of the file:
Snapshot of for . Not updated: later research supersedes it.
Without a ticket, the line drops the ticket: "Snapshot of . Not updated: later research supersedes it."
-
Save it where the repo already keeps such notes, matching the existing convention; if it keeps none, in
docs/research/. -
If an earlier report answers the same question, this one supersedes it: add
Superseded by <relative link to the new report>.right after that report's snapshot line. Once a report is onmain, that line is the only edit it ever takes. -
Return the report's path, a one-line gist of its answer, and the path of the report it supersedes, if any.
The agent reads anything in the checkout and runs read-only commands, and leaves the report in place, uncommitted, for the session that launched it: it writes nothing to git, no branch, commit or push.