v2.1.0 - Evidence contract and workflow reliability

Evidence Contract and Workflow Reliability

Fixes the verified evidence-gate, initialization, installation, and provider-routing defects from the September audit.

• Factual claims need resolvable, reviewed evidence records. Missing sources, planned results, broken IDs, absent artifacts, and nonexistent inputs cannot pass.
• Explicit limitations, assumptions, proposals, and excluded backlog claims remain distinct from factual assertions. Section-scoped checks are supported.
• Stage gates share the same contracts, require applicable artifacts and recorded human review, and no longer confuse structure checks with completion.
• All three project initializers persist their type. Thesis delivery checks requirement IDs and evidence mappings; reviewed non-quantitative analysis exemptions are supported.
• Provider checks cover metadata, linked resources, companions, explicit roots, and Codex/Claude discovery paths. Discovery is not execution approval. Nested test skills are no longer discoverable.
• Demo replacement protects unowned directories and user additions. Dependencies, cold-start instructions, and cross-platform checks are included.

Original modes, internal modules, 18 optional provider mappings, artwork, and bilingual README ordering are preserved.

Validation

45 regression tests, 30 CSV records/templates, both demo workflows, generated-output checks, and skill/README structure checks pass. GitHub CI passes on Windows and Ubuntu with Python 3.10 and 3.12.

CI results

Existing Workspaces

Workspace schema version 2 adds required evidence and review records. Existing CLI names remain available, but permissive historical readiness results are not preserved. Preview migration first; --apply backs up originals and never promotes evidence status. Research workspaces are not automatically migrated by installing this release.

Migration guide · Runtime setup · Evidence contract · Audit remediation scope

Offline checks establish recorded consistency, not scientific truth or genuine reviewer identity. No end-to-end performance claim is made for all third-party providers.

v2.0.0 - Capability-routed research workflow

Academic Paper Writing Skill v2.0.0

This release upgrades the project from a strong evidence-first writing workflow into a capability-routed research system.

Highlights

• 13 stable research capability interfaces with 18 optional specialist provider mappings
• Internal fallbacks for ideation, scholarly search, study design/statistics/data, submission, and presentation
• Full provider detail when installed, with normalized handoffs and output acceptance
• Stage gates for ideation, literature, analysis, drafting, submission, and presentation
• Preserved original project types, modes, modules, demos, and pre-prose guardrails
• Rebuilt bilingual README with the retired large hero image removed

See capability_registry.yaml and docs/CAPABILITY_COVERAGE.md for the interface design.

v0.1.0 - Evidence-first academic writing workflow

What this release does

• Builds literature matrices before related-work prose.
• Maps claims to verified sources and experiment evidence.
• Separates planned, preliminary, and completed results.
• Supports research papers, literature reviews, rebuttals, and undergraduate thesis workflows.
• Adds issue templates, PR template, filled sample outputs, and social-preview copy.

What it does not do

• It does not fabricate papers.
• It does not fabricate SOTA.
• It does not fabricate results.
• It does not invent school, advisor, rubric, template, defense, or workload requirements.

Try it in 3 minutes

python scripts/init_project.py --out paper_workspace --type research_paper
python scripts/validate_skill.py

See examples/outputs/ for filled sample matrices, claim ledgers, novelty checks, and thesis evidence maps.