opl-connect-source-and-skill-router

name

opl-connect-source-and-skill-router

description

Use for OPL Connect external skill/source search-inspect-sync routing, connector receipt debugging, single-skill sync decisions, and source refs/no-authority review. Helps Codex decide whether an external source or skill is needed, which candidate to inspect or sync, whether connector metadata or receipts are too risky, and how to hand candidate refs to a domain owner without claiming domain truth, quality verdict, owner acceptance, or readiness.

OPL Connect Source And Skill Router

Use this skill when a task needs OPL Connect to find, inspect, debug, or selectively sync an external source, connector, package descriptor, or Codex Skill.

Optional helper: kernel.py provides deterministic source/ref normalization, request classification, handoff skeleton, and forbidden-claim lint helpers. It is stdlib-only, writes nothing, performs no network or subprocess calls, and does not claim domain truth, owner acceptance, or readiness.

Boundary

• Treat OPL Connect as the program owner for registry lookup, source and skill search, candidate inspection, selective sync, API normalization, and connector invocation receipt candidates.
• Treat this Skill as the AI judgment layer: decide whether external source or skill help is needed, which candidate fits the task and owner, whether the risk is too high, whether exactly one Skill should be synced, and how results should be handed to the owning domain.
• Keep domain truth with the domain owner. Connect output and this Skill's judgment are candidate refs only.
• Treat approved source registries, including K-Dense-AI/scientific-agent-skills when registered as kdense-scientific-agent-skills, as searchable source inventories, not default task context or default Codex metadata.
• Treat external specialist routing as gap-gated: only search/inspect external registries after the active default OPL/domain pack is insufficient, and only sync one selected Skill to workspace/quest scope when the task actually needs local instructions.

AI-first / Contract-light Semantics

• Use Connect contracts and modules only for source identity, capability kind, candidate refs, sync receipts, recovery, and verification.
• Keep elastic routing judgment in this Skill: decide whether a source or Skill is needed, inspect the smallest candidate, weigh authority risk, and shape route-back.
• If metadata is ambiguous or stale, return a refs-only hold or owner route instead of broadening default exposure or hardcoding source semantics.
• Keep K-Dense/scientific-agent-skills and similar libraries as registered registries. Do not translate registration into default install, bulk metadata, global/user Codex exposure, or a new alias router.

Router Workflow

1. Classify the request as source_search, skill_search, candidate_inspect, single_skill_sync, connector_receipt_debug, or refs_only_review.
2. Start with search or an explicit selector only after naming the default-pack gap. Use list/index views only for maintainer source review; do not turn a whole external library or its full metadata set into default context.
3. Inspect the smallest plausible candidate before syncing. Check owner, source path or URL, capability kind, sync scope, expected target, authority boundary, and stale or missing source signals.
4. Prefer one-skill sync. Sync only the selected Skill or refs-only subset needed for the current workspace or quest unless the domain owner explicitly requires a package-level install.
5. For connector receipt debug, compare requested resource, normalized refs, invocation parameters, errors, no-authority flags, and receipt candidate path. Classify failures as access, normalization, receipt construction, sync target mismatch, or authority overclaim.
6. Hand off source_refs, candidate_refs, sync_receipt_ref, connector_invocation_ref, no_authority_flags, owner_route, residual risk, and the next legal owner action.

Forbidden Claims

• Do not full-install external skill libraries, source repos, runtime packages, caches, generated assets, or connector payloads by default.
• Do not expose approved external source metadata broadly just because a source is registered; registration allows search, inspect, and selected sync only.
• Do not treat connector metadata, registry entries, descriptor status, sync success, cache existence, package payload presence, or invocation receipts as domain truth, quality verdict, owner acceptance, artifact authority, typed blocker, human gate, release readiness, production readiness, or domain readiness.
• Do not write domain truth, owner receipts, typed blockers, runtime queues, provider attempts, artifact bodies, quality verdicts, or live readiness surfaces from this Skill.
• Do not promote a contract module, source index, or reference pack into a real Codex Skill unless there is a true stage prompt or professional Skill wrapper with its authority boundary stated.