Operate the docs/research compiled layer in either full or maintain mode. Use for research-lane creation, research upkeep, authority-sensitive corpus work, source summaries, concepts, systems, decisions, and open-question pages.
Handle $ARGUMENTS. This is the reusable workflow for operating the
docs/research compiled layer without turning research into one-off chat mush.
<task>#$ARGUMENTS</task>
Turn the existing research command docs into one reusable skill entrypoint.
This skill should decide whether the job is:
The support docs remain canonical detail:
Use them as sub-workflows, not as dead docs.
docs/research../rawfull mode, verify corpus completeness via an official-source discovery
step before declaring a corpus missing, thin, or only partially evidenced.full mode, once official-source discovery confirms that the relevant raw
family already exists locally, exhaust the strongest local raw hits before
calling evidence gap or contradiction gap.full mode, every scoped corpus must end in an explicit disposition:
open-questions/ page instead of flattening uncertainty into fake law.Always read:
Then choose one support doc:
fullmaintainfull when:Examples:
maintain when:docs/researchWhen full is chosen for a surface that obviously touches multiple editors or
source families, do not stop after finding one useful source.
Do the full corpus-level pass:
index.md and log.mdFor editor-behavior authority work, that often means some subset of:
depending on the actual surface
For full mode, explicitly close each scoped corpus with:
evidencedraw gapcompile gapsynthesis gapfreshness gapevidence gapcontradiction gapstructure gapDo not end a full pass with silent corpora.
If a likely corpus was scoped and produced no useful proof, say that directly. That is still a result.
Before recording raw gap, use official-source discovery to confirm that the
missing coverage is real and not just a stale or incomplete local raw mirror.
Before recording evidence gap, read the strongest local raw hits that your
searches surfaced for that corpus. Broad grep is routing, not proof.
If raw evidence is missing for a scoped corpus after that discovery step,
record a raw gap explicitly instead of quietly letting another corpus carry
the whole answer.
If compiled pages are missing but raw exists, record a compile gap and create
the missing source summary when safe.
If evidence exists but still does not answer the behavior question, record an
evidence gap or contradiction gap instead of flattening it into a partial
yes.
If official discovery and local raw search already surfaced an obviously relevant file, you do not get to stop at "broad grep over corpus X". Read the file and cite it, or explain why it was not actually relevant.
A successful pass should leave behind durable artifacts such as:
full if you only spot-checked one corpus when the
topic obviously spans several.full mode without an official-source discovery step for each
scoped corpus.full mode without a per-corpus evidence ledger.index.md or log.md stale after adding research pages.