sr-survey-prior-router

name	sr-survey-prior-router
description	Draft/staged knowledge-resource router for prepared research resources, domain prior, literature corpora, survey/SR writing prior, evidence grounding, source audit, and source-grounded synthesis. Use to choose the smallest relevant reference knowledge; this draft does not execute workflows or mutate runtime state.

SR/Survey Prior Router

This is a draft/staged knowledge-resource router that can be read in place from a standalone checkout, staged inside another repository, or installed into an agent skill root. Interpret referenced paths relative to the directory containing this SKILL.md. Installation or runtime activation is a separate, explicit action; once installed, agents/openai.yaml controls whether Codex may also invoke it implicitly from the skill description.

Its job is navigation: point the agent to the smallest relevant reference knowledge and keep source/provenance boundaries visible. It is not an execution workflow, automation harness, or multi-step prompt pack.

Coverage Status

This pack now includes a seed source inventory under references/source_inventory/, a first local full-document corpus under references/canonical_sources/, and lazy-load locator indexes under references/corpus_index/. Treat the corpus as the local read path for downloaded official/method sources, not as proof of exhaustive coverage: broad APIs, commercial sources, blocked URLs, and source-specific licensing still need route-by-route checks.

Before claiming substantive survey-writing or SR-writing coverage, consult references/source_inventory/coverage_report.md, verify the relevant source_manifest.jsonl / source_registry.yaml rows, then use references/route-source-index.yaml, references/source_inventory/local_corpus_index.*, and references/corpus_index/ to locate the smallest relevant local Markdown section. Prefer local canonical_sources/md/<source_id>/ files only after this source/document/section selection is done. Split corpus_index/sections/by_source/<source_id>.jsonl files may exist locally as generated backing locators but are not part of the ordinary GitHub review surface; if absent, fall back to corpus_index/document_index.jsonl, source cards v2 locators, and targeted search inside the selected canonical Markdown file. If a graphify graph exists for this pack, use references/graphify-navigation.md only as an optional navigation adapter after route selection; graph output is never canonical evidence.

Use When

• The user asks how prepared research resources, survey-writing material, SR-writing resources, or broader domain material should become agent prior.
• The task needs knowledge routing across a literature corpus, source registry, evidence packet, citation audit, retrieval layer, or source-grounded synthesis question.
• The task must keep long knowledge out of the prompt while still making the agent read the right prepared source.

Do Not Use When

• The task is ordinary coding, shell usage, or writing unrelated to prepared research resources.
• The user asks for a current fact that needs live web or a specific external source not represented in the prepared corpus.
• The user asks to install, activate, or mutate an agent runtime. This staged pack only describes behavior.

Reference Navigation

1. Read references/route-registry.md first.
2. Select exactly one smallest route unless the task clearly crosses route boundaries.
3. Read only the reference files listed for that route.
4. If the route needs local full-document evidence, read references/route-source-index.yaml, then choose source IDs through references/source_inventory/local_corpus_index.md or .json.
5. Use references/corpus_index/document_index.jsonl and, when available locally, the selected references/corpus_index/sections/by_source/<source_id>.jsonl file to find line/page locators before opening any canonical Markdown file. If split section locators are unavailable in a public checkout, use references/source_inventory/source_cards_v2/<source_id>.md and targeted search in the selected canonical Markdown file as the fallback.
6. Use graphify only when the selected route still needs relationship navigation or candidate narrowing. Follow references/graphify-navigation.md, then return to document/section locators and canonical verification.
7. Consult optional notes only when they clarify a knowledge distinction.
8. If a source, registry, locator, graph, or corpus is missing, say what is missing instead of guessing.

Safety Rules

• Do not modify frozen SRs: 2307.05527, 2409.13738, 2511.13936, or 2601.19926.
• Do not bulk-load references/canonical_sources/raw/, references/canonical_sources/md/, or every corpus_index/sections/by_source/*.jsonl file.
• Do not treat compiled wiki, RAG output, graphify output, source cards, indexes, or summaries as canonical truth.
• Do not present unsupported claims as conclusions.
• Keep repo rules, source provenance, and read/write boundaries visible in the answer.

Answer Guidance

For audit-heavy or evidence-sensitive answers, make these distinctions visible when they matter:

• read: files, registries, or sources actually consulted.
• supported: claims or actions supported by the consulted sources.
• uncertain: missing evidence, weak support, stale information, or unresolved routing.
• next_smallest_action: the next bounded action that would reduce uncertainty.

Do not force this shape for simple answers. Use it only when it helps preserve provenance or uncertainty.