Academic Pipeline Orchestrator
From research to publication: a 10-stage academic paper workflow with mandatory integrity verification and two-stage peer review.
Evidence shows an explicit least-privilege philosophy and user confirmation mechanisms, such as mandatory checkpoints and user confirmation. However, permission control relies on environment variables and external tools like WebSearch and API calls, with no explicit sandbox or isolation measures. Data-flow disclosure is incomplete, especially external API calls and caching mechanisms. Dependency security is not stated. Deductions: incomplete permissions and isolation, insufficient disclosure of external effects and security handling.
Static review shows good instruction self-consistency, with clear state machine and process definitions, but key paths (like citation verification and retrieval) depend on external services and tools, with no reproducible test evidence. Error handling for abnormal inputs exists (e.g., NOT_FOUND, MISMATCH) but its effectiveness and feedback quality are not verified. Deductions: lack of execution verification, unverified error-handling feedback quality.
Scope is clear, with clear trigger scenarios and non-trigger scenario tables, and detailed semantic trigger keywords including Chinese. However, core functionality depends on overseas services (Semantic Scholar API, WebSearch), which may be unreachable in mainland China, with no alternative or degradation strategy provided. Deductions: insufficient environmental fit, especially mainland China network reachability.
Documentation structure is hierarchical, with version numbers, update dates, and metadata, but lacks clear installation/dependency notes, troubleshooting guides, known-limitation disclosures, changelogs, and license clarity. Dependency documentation exists but is not independently detailed. Deductions: incomplete governance documentation, especially license and changelog absence.
Value proposition is clear, goals are defined, but static review cannot verify actual output quality. Process design is detailed, but marginal value is not compared to manual or alternative methods. Deductions: no verifiable output evidence, cost/benefit ratio not assessed.
Files contain many internal references and design docs, but lack independent third-party verification evidence. No reproducible test results or CI logs provided. Deductions: single evidence type, lack of cross-source corroboration.
- Core functionality relies on overseas services like Semantic Scholar API and WebSearch, which may be unreachable in mainland China; consider alternatives.
- No test or CI evidence provided; static review cannot verify actual runtime reliability; check repository for test files.
- License is NOASSERTION; confirm copyright and compliance risks before use.
- Depends on multiple external skills (deep-research, etc.); ensure they are reliable and compatible.
What does this skill do, and when should you use it?
academic-pipeline is a lightweight orchestrator that manages the complete academic workflow from research exploration to final manuscript. It performs no substantive work itself but detects stages, recommends modes, dispatches sub-skills (deep-research, academic-paper, academic-paper-reviewer), and tracks state with mandatory user confirmation checkpoints. Its core differentiator is the enforced integrity verification (Stages 2.5 and 4.5) and two-stage review (Stages 3 and 3') to ensure citations and data are 100% accurate. The skill supports mid-entry from any stage and optional cross-session resume via a Material Passport.
Detects the user's current stage (research, write, integrity check, review, revise, etc.) and available materials; recommends optimal modes per stage (e.g. socratic, full, quick); dispatches the corresponding sub-skill for each stage; requires user confirmation after each stage completion; runs integrity_verification_agent to check references, data, and claims; manages two review-revision loops; and produces a process record (Paper Creation Process Record) with a progress dashboard.
- Starting from scratch: “I want to write a research paper on the impact of AI on higher education quality assurance” launches the full pipeline from Stage 1.
- Have a draft: “I already have a paper, help me review it” enters mid-pipeline at Stage 2.5 for integrity checks before review.
- Received reviewer comments: “I received reviewer comments, help me revise” starts at Stage 4 to address feedback.
- Cross-session resume: In a long workflow, use `resume_from_passport=<hash>` in a fresh session to continue from a recorded stage.
- Ensuring integrity: Run integrity verification on an existing paper to validate citations and data.
What are this skill's strengths and limitations?
- Enforced integrity verification reduces citation hallucination risk
- Two-stage review ensures revision quality
- Flexible mid-entry from any stage
- Detailed process record and progress dashboard
- Mandatory user checkpoints prevent silent errors
- Relies on Claude Code-specific features (e.g. subagents), requiring adaptation for other platforms
- Requires extra tools like Pandoc/tectonic for DOCX/PDF output
- Data access level is verified_only, limiting raw data operations
- Steep learning curve due to extensive documentation
How do you install this skill?
In Claude Code, install via the plugin marketplace: /plugin marketplace add Imbad0202/academic-research-skills then /plugin install academic-research-skills. Alternatively, clone the repository and place the academic-pipeline folder into Claude Code's skills directory (see docs/SETUP.md for paths). Requires Claude Code and an ANTHROPIC_API_KEY.
How do you use this skill?
Trigger with a natural-language prompt in Claude Code, e.g. “I want to write a research paper on the impact of AI on higher education quality assurance” to start the full pipeline. Or provide partial materials for mid-entry. During the flow, wait for and confirm each stage checkpoint. You can check progress with “status”.