Ref Query

ℹ️ First, set expectations: before anything else, print one short line with this skill's version and its most recent change — the top entry of metadata.changelog above — e.g. ref-query v1.0 — recent: initial version. Then proceed.

Answer a structured question about reference records with the temporal contract applied. This is not /recall — /recall searches the user's cognitive insights; ref-query reads the CRM/market source of truth and honours status/validity/as_of.

Read first: resources/layered-brains/COMPANY-BRAIN-SCHEMA.md → "The temporal model" (read-time interpretation + the per-type Freshness SLA). This skill enforces those rules.

State Dependencies

Process

Step 1 — Classify the question

• Entity lookup ("what do we know about X as of today?") → resolve the note (Grep by name/alias, or a mounted search) and read it.
• Filtered list (by type/relationship/status/expiry window) → deterministic frontmatter scan (below), not semantic search — the answer must be exact and reproducible.
• Fuzzy / exploratory ("who might overlap with X?") → mounted semantic search:

  BRAIN_READ_SCOPE=core,Company resources/local-brain-search/run_search.sh "<query>" --limit 10 --json

Step 2 — Scan (deterministic filters)

Frontmatter is the query surface. Examples:

cd Brain/Company
# all active competitors
grep -rlE '^relationship: competitor' */*.md | while read f; do grep -qE '^status: active' "$f" && echo "$f"; done
# active engagements + their as_of
for f in engagements/*.md; do grep -qE '^status: active' "$f" && printf "%s  as_of=%s valid_until=%s\n" \
  "$(basename "$f" .md)" "$(grep -m1 '^as_of:' "$f" | cut -d' ' -f2)" "$(grep -m1 '^valid_until:' "$f" | cut -d' ' -f2)"; done

For an "expiring in N days" filter, compare valid_until to today + N. Note: notes that predate incremental valid_until adoption have no window — report them as "no recorded term" rather than dropping them.

Step 3 — Apply read-time interpretation (mandatory)

Before presenting any fact:

1. Prefer status: active over superseded (a superseded note is history — show it only if asked "what changed?").
2. A fact past its valid_until is stale-unless-renewed — say so.
3. Discount by as_of age against the per-type Freshness SLA (COMPANY-BRAIN-SCHEMA): market product/pricing ~14d · engagement ~30d · client organization ~90d · person ~180d · own product on-change. Over SLA → tag ⚠ stale (as_of N days old).
4. Always print the as_of date next to each fact ("Trinity pricing as of 2026-06-13").

Step 4 — Answer

Compact, sourced, staleness-legible. One line per record: name — key fact (as_of YYYY-MM-DD [· ⚠ stale] · status). For an entity lookup, summarise the body + list its [[relations]] + the freshness verdict.

Verification

• A real entity question ("what do we know about Acme Corp as of today?") returns the note's facts with the as_of date printed and a freshness verdict.
• A real filter ("which engagements are active?") returns exactly the notes whose frontmatter matches (cross-check the count against grep -rlE '^status: active' engagements/*.md | wc -l).
• Nothing is written; git status is unchanged after a query.

This skill must refuse to

• Write or mutate any note (read-only).
• Present a superseded fact as current, or a fact without its as_of date.
• Set/alter provenance, crystallize, lifecycle-classify, or promote a record into a cognitive insight.