YCC-Lover/research-starter-skills / skills/research-starter-paper
research-starter-paper
Unverified科研学习与论文写作总入口,协调六个模块。帮助不知道提供什么材料的新手逐步产出、自动保存和恢复项目进度,也用于跨阶段研究任务;区分学习、执行和只读检查。
resources/research-engine/academic-paper/references/workflow_phase_details.md
View on GitHub# Orchestration Workflow — Phase Details
Detailed per-phase agent behavior and output descriptions for the 8-phase orchestration workflow.
---
## Phase 0: CONFIG (Interactive)
**Agent**: `intake_agent`
**Output**: Paper Configuration Record
- Paper type (IMRaD / Lit Review / Theoretical / Case Study / Policy Brief / Conference)
- Discipline and sub-field
- Target journal (optional)
- Citation format (APA 7 / Chicago / MLA / IEEE / Vancouver)
- Output format (LaTeX / DOCX / PDF / Markdown / Combined)
- Language (EN / zh-TW / bilingual sections)
- Bilingual abstract (Yes / EN-only / zh-TW-only) — cardinality only; it does not select which languages
- Output language pair (`output_language_pair`) — the two abstract languages, from the registry in `shared/output_language_pair.md`; default `zh-tw-en`, per-run override; the PCR row is omitted when the run declares none
- Word count target
- Existing materials (RQ, data, drafts, lit)
**Checkpoint**: User confirms configuration.
---
## Phase 1: RESEARCH
**Agent**: `literature_strategist_agent`
**Output**: Search Strategy + Source Corpus
- Database selection + search strings
- Inclusion/exclusion criteria
- Source screening + annotated bibliography
- Literature matrix (Source x Theme)
- Research gap mapping
**Checkpoint**: User reviews sources (optional add/remove).
---
## Phase 2: ARCHITECTURE
**Agent**: `structure_architect_agent`
**Output**: Paper Outline + Evidence Map
- Structure pattern selection (from paper_structure_patterns.md)
- Section-by-section outline with word count allocation
- Evidence-to-section assignment
- Transition logic between sections
**Checkpoint**: User approves outline.
---
## Phase 3: ARGUMENTATION
**Agent**: `argument_builder_agent`
**Output**: Argument Blueprint
- Central thesis + sub-arguments
- Claim-Evidence-Reasoning chains per section
- Counter-argument identification + rebuttal strategy
- Logical flow diagram
---
## Phase 4: DRAFTING
**Agent**: `draft_writer_agent`
**Output**: Complete Draft
- Section-by-section writing following outline
- Register adjustment for discipline
- In-text citations integrated
- Word count tracking per section
- Transition paragraphs between sections
- Acronym check on the saved draft (#849; [`writing_quality_check.md`](writing_quality_check.md) § F)
---
## Phase 5a & 5b: CITATIONS + ABSTRACT (Parallel)
### Phase 5a: Citations
**Agent**: `citation_compliance_agent`
**Output**: Citation Audit Report
- In-text <-> reference list cross-check (zero orphans)
- Format compliance (per selected style)
- DOI/URL verification
- Self-citation ratio check
- Auto-correction of detected errors
### Phase 5b: Abstract
**Agent**: `abstract_bilingual_agent`
**Output**: Bilingual Abstract + Keywords (both languages of the run's declared output language pair)
- L2 abstract (default: English), structured — length from the regime table ([`abstract_writing_guide.md`](abstract_writing_guide.md), run's paper type)
- L1 abstract (default: Traditional Chinese), structured — length from the same regime table
- L2 keywords, count from the same regime table
- L1 keywords, same declared count
- Headings and report columns are pair-derived; for the default pair `zh-tw-en` they render exactly as `### English Abstract` / `### Chinese Abstract`
- Independent writing (not mechanical translation)
- Field absent → default pair; the heading literals are unchanged from the pre-#862 surface (no rendered-output equivalence is claimed)
- Acronym check on the saved abstracts (#849; [`writing_quality_check.md`](writing_quality_check.md) § F)
---
## Phase 6: PEER REVIEW
**Agent**: `peer_reviewer_agent`
**Output**: Review Report + Revision Instructions
- 5-dimension criterion-bound judgement:
Originality | Methodological Rigor | Evidence Sufficiency |
Argument Coherence | Writing Quality
- Every live Phase 6 report and downstream review package emits
`calibration_status: NOT_CALIBRATED` unconditionally in the current release.
Candidate empirical target profiles declare
`application_status: NOT_WIRED_TO_LIVE_REVIEW`; attaching one or observing an
apparent metadata match cannot upgrade the live status. No weighted total or
fixed verdict mapping is available.
- Verdict: Accept / Minor Revision / Major Revision / Reject
- Line-level feedback with suggested fixes
- Max 2 revision loops -> back to Phase 4 [draft_writer_agent] (limited to 1 round in academic-pipeline)
---
## Phase 7: FORMAT
**Agent**: `formatter_agent`
**Output**: Final Output Package
- Target format conversion (LaTeX + .bib / DOCX / PDF / Markdown)
- Journal-specific formatting (if target journal specified)
- Cover letter (if journal submission)
- AI disclosure statement
- Final quality checklist
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |
| fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks. | Stars: 115k | Updated | Claude CodeCodex | Verified | Install |