search — Unified Search Hub

Search discipline for any lookup that leaves the repository. This skill is implicit-visible as metadata (allow_implicit_invocation: true); load the full body on explicit trigger or dev-hub routing, never by an external dispatcher.

Source-Proof Invariant (read first)

Search results are candidate URLs, not evidence. Snippets, summaries, and search-result consensus discover where a fact might live; they never settle it. When recency, factual accuracy, version/compatibility, or source attribution matters, open the original page and confirm it before you treat the answer as sufficient. This invariant precedes every sufficiency rule below — no tier may declare an answer final on snippet text alone.

The Ladder (exactly three codex-native tiers)

Research-depth classifier (SEARCH-DEPTH-01)

Before climbing the ladder, name the depth — it is distinct from the target classifier in the Korean Intent Guard (which picks web vs docs vs repo):

• latest/current fact — one entity, a version/date/price/status. Tier 1 discover + Tier 2 open one primary source. Capture the exact date.
• official-doc fact — API/library behavior. Prefer official docs first, then open for proof.
• implementation/source fact — how something is built. Open the source/repo, not a summary.
• comprehensive research — multi-source/contested. This is the only depth that justifies Tier 3 (Deep research); ordinary latest/current lookups never auto-escalate to a subagent swarm.

Target routing

Classify external/public/current information, official-doc behavior, or this repository's code/logs/config before choosing tools; system comprehension routes to cxc-dev-debugging. Before querying, read Korean Intent Guard and query rewrite.

Conditional References

Condition	Reference
Before every query	Korean Intent Guard, query rewrite
Discovering candidate URLs (Tier 1)	Hosted discovery
Opening a source to settle a claim (Tier 2)	Source-open proof (SEARCH-BROWSE-01)
Blocked, JS, PDF or table source	Blocked-URL reader
Explicit deep-research request or host mode (Tier 3)	Entry boundaries, then Deep research
Any search subagent, including audit/reviewer or Luna lanes	Skill attachment (SEARCH-ATTACH-01)
PABCD divergence candidates	Candidate grounding
Luna discovery or assessing available backends	Discovery boundaries
Host/tool capability or search integration questions	Tool notes

When to stop

Stop escalating when any holds: a sufficient primary source is found and confirmed; all candidate URLs are dead or unreachable; or the task needs a user clarification. Do not keep climbing tiers past a confirmed answer, and do not spend Tier 3 subagents on a question Tier 1+2 already settled.