Academic Advisor

Academic Advisor

Provide advisor-style evaluation for proposals, rough ideas, research plans, manuscripts, and journal targeting decisions. The output is one integrated Simplified Chinese report, not separate sub-skill reports.

Defaults

• Read ../shared/research-calibration.md and infer the research stage, contribution level, rigor profile, and compute profile before judging the project. If the user does not specify them, use the shared defaults.
• Read ../shared/advantage-led-research-narrative.md. Lead the assessment with the project's strongest honest advantage, the arena in which it can matter, and the decisive evidence needed to establish it. Keep speculative reviewer attacks out of the headline verdict.
• Ground decisive recommendations in evidence: read references/zotero-evidence-workflow.md, check Zotero readiness, and search Zotero first. For rough ideas and first drafts, use a targeted search around the decision-critical claims; reserve comprehensive novelty and journal benchmarking for submission or explicit deep-evaluation requests.
• Default report language: Simplified Chinese. English manuscripts, English journal names, and English source metadata do not change the report language. Use English only when the user explicitly asks for an English report or when a specific submission-ready English passage is requested.
• Default domain baseline: geography, human geography, GIScience, health geography, spatial analysis, and environmental or health geography when relevant.
• Match journal ambition to the contribution and evidence already present. Include a stretch route only when its minimum upgrade is plausible; do not treat specialist, regional, replication, or incremental venues as inferior when they are the best scholarly fit.
• Before naming or ranking journals, read references/local-journal-reference.md and attempt to load the user's local journal_ai_reference.md. Use it as the default candidate pool and preference ledger, then verify current scope, guidelines, and unstable metrics from primary journal or publisher sources.
• Do not edit the proposal or manuscript by default. Produce a separate assessment report unless the user explicitly asks for rewriting.

Use the Research Continuity record in ../shared/advantage-led-research-narrative.md: preserve question, advantage state, fair comparison, claim/experiment IDs, evidence version, and retellable sentence across stages. Update only fields affected by the current task; reuse project records rather than starting another ledger.

Routing Boundaries

• Use academic-advisor for author-side project diagnosis, pre-submission audit, and journal strategy.
• Use good-question logic for question quality, advantage hypothesis, falsifiability, and pilot design. Add reviewer rejection analysis only for pre-submission, revision, reviewer/grant-like, or materially risky cases.
• Use good-story logic for completed manuscripts whose evidence, figure order, central claim, or narrative spine needs diagnosis.
• Use academic-research logic for literature positioning, research design, manuscript structure, contribution, and its internal paper or reviewer workflows.
• Use zotero:Zotero for local-library evidence search, citation metadata, and full text only when needed. If the plugin skill is not automatically loaded, follow references/zotero-evidence-workflow.md to locate and run the current Zotero helper without hard-coding a cache hash.
• Use peer-review instead for formal referee reports, external manuscript review, or journal-style reviewer comments not tied to the user's author strategy.

Required Workflow

1. Intake the provided proposal, idea file, manuscript, and optional journal list.
2. Classify the task as idea-proposal, manuscript-pre-submission, journal-targeting-only, or revision-strategy.
3. Apply the shared calibration profile:
   • Name the research stage and contribution level.
   • Separate checks needed now from checks that can be deferred.
   • Do not penalize a first draft for omitting submission-stage robustness unless the omission makes the current result invalid or misleading.
4. Build evidence before giving a mature recommendation:
   • Read references/zotero-evidence-workflow.md.
   • Run a Zotero readiness check and search Zotero for core theory, methods, adjacent papers, target-journal examples, and missing references.
   • Scale search depth to the decision: targeted evidence is enough for an early-stage diagnosis; journal scope, current guidelines, recent special issues, and representative articles require current web verification when recommending a submission route.
   • Maintain an evidence ledger with 来源支持 Source-backed, 推断 Inference, and 未知/待核查 Unknown.
   • If Zotero is unavailable, do not silently skip it; report the exact blocker in the evidence ledger.
5. Select only the lenses needed for the classified task:
   • idea-proposal: good-question for importance, tractability, advantage hypothesis, fair comparison, falsifier, pilot, and material assumptions; academic-research for literature positioning, design, methods, and contribution.
   • manuscript-pre-submission: good-story for the central claim, evidence ladder, leading advantage, winning arena, experiment jobs, and figure order; academic-research for structure, methods, claim-evidence boundaries, and readiness.
   • journal-targeting-only: journal-fit and evidence-triangulation rubrics; add good-story only when manuscript materials are available and story shape affects fit.
   • revision-strategy: load only the question, story, research, or journal lens implicated by the requested revision.
6. For every journal-targeting task:
   • Read references/local-journal-reference.md and run its path-resolution sequence before recommending journals. Prefer python scripts/resolve_journal_reference.py --project-root <project>; pass .codex/project.yaml's paths.journal_reference value through --configured when present.
   • Record whether the local reference was loaded, its visible update date, and which resolver found it. Do not silently skip an unreadable file.
   • If the user supplies another journal list, treat that list as the active constraint; use the local reference to enrich it, not replace it.
   • Read references/journal-fit-rubric.md and verify current external facts that could change the recommendation.
7. Produce one integrated report using templates/integrated-advisor-report.md.

Reference Loading

• Read references/advisor-rubric.md for proposal, idea, manuscript, and revision evaluation.
• Read references/evidence-triangulation.md before making novelty, literature-gap, or target-journal claims.
• Read references/zotero-evidence-workflow.md whenever evidence grounding is required.
• Read references/journal-fit-rubric.md when the user provides a journal list or asks where to submit.
• Read references/local-journal-reference.md before every concrete journal recommendation or ranking.
• Read ../shared/research-calibration.md for stage, rigor, contribution, and compute defaults.
• Read ../shared/advantage-led-research-narrative.md for author-side emphasis, experiment purpose, winning-arena selection, and materiality triage.
• For a completed manuscript whose story affects submission readiness, read ../good-story/SKILL.md and preserve its stable story handoff.
• For geography-related proposals or manuscripts, read ../shared/field-context/geography-publication-readiness.md and include its five-gate diagnosis.

Output Rules

• Lead with a Chinese bottom-line verdict, the paper's leading advantage, its strongest credible route, and the evidence needed to make that route work.
• Separate source-backed claims from inferences and unknowns using the Chinese evidence ledger labels.
• Do not output the final integrated report in full English unless the user explicitly requests English.
• Do not claim "nobody has studied this" unless the evidence search supports that claim.
• State the contribution level without demanding category uniqueness. Replication, regional extension, measurement improvement, and modest method gains are viable when their value and venue fit are explicit.
• For manuscripts, inspect actual text structure and, when figures or tables carry claims, use pdf or docx visual/document tools as appropriate.
• For journal ladders, lead with the strongest credible route. Include stretch target, realistic target, and fallback target when all three are useful; omit a performative stretch route when no plausible upgrade path exists.
• Keep the final report directly reusable as a project document.
• Do not make a weakest-gate or reviewer-attack inventory mandatory. Surface a constraint only when it changes the central claim, development plan, or journal route.