v0.3.0 - Clean knowledge-base system, setup skill, and usage flow

v0.3.0 - Clean knowledge-base system, setup skill, and usage flow

Self-Learning Library v0.3.0 publishes the cleaned knowledge-base system as a reusable package. The release focuses on portability, skill-based setup, clearer rule boundaries, and a documented end-to-end usage flow.

What is upgraded

• Added a portable Codex skill under skills/paper-kb-project-setup.
• Added optional deployment of bundled generalized paper-writing rules during setup.
• Added the 45_Supervision layer as an explicit setup target for user corrections, mentor corrections, regression findings, and high-priority constraints.
• Updated PROJECT_HARNESS_WORKFLOW.md with the non-source document Markdown extraction guard.
• Refreshed 40_Final_Generalized_Rules to keep only broadly reusable writing rules.
• Moved the interface-wave/intermetallic-compound dual-mainline rule out of final generalized rules and into 60_Limited_Rules.
• Added a dedicated v0.3 usage flow in docs/usage-flow-v0.3.0.md.
• Kept generated full-evidence JSON outputs out of the public package.

Why this release matters

Earlier releases demonstrated the layered vault and local RAG loop. v0.3.0 makes the system easier to redeploy into a new project: install the setup skill, choose whether to seed generalized paper rules, open the generated vault in Obsidian, then use local RAG and iteration from the target workspace.

The key boundary is stricter in this release:

• reusable paper-writing rules stay in 40_Final_Generalized_Rules;
• project-only or manuscript-specific rules stay in 60_Limited_Rules;
• workflow/process rules stay in 35_Workflow_Governance;
• active corrections and regression constraints stay in 45_Supervision.

Quick verification

The release should pass:

py -m py_compile tools\kb_rag.py tools\paper_iteration.py skills\paper-kb-project-setup\scripts\setup_paper_kb_project.py
py tools\kb_rag.py --query "SCI abstract topic purpose method result conclusion" --limit 1
powershell -NoProfile -ExecutionPolicy Bypass -File .\run_paper_iteration.ps1

To verify the setup skill, run it against a disposable workspace:

py skills\paper-kb-project-setup\scripts\setup_paper_kb_project.py --target "C:\temp\paper-kb-v030-smoke" --with-paper-rules
py "C:\temp\paper-kb-v030-smoke\tools\kb_rag.py" --query "workflow supervision limited rules" --limit 3
py "C:\temp\paper-kb-v030-smoke\tools\paper_iteration.py" --root "C:\temp\paper-kb-v030-smoke"

Upgrade guidance

For an existing clone:

git pull
py tools\kb_rag.py --query "your task keywords" --limit 3
powershell -NoProfile -ExecutionPolicy Bypass -File .\run_paper_iteration.ps1

For a new project, install or copy the setup skill and initialize a target workspace:

py skills\paper-kb-project-setup\scripts\setup_paper_kb_project.py --target "C:\path\to\workspace" --with-paper-rules

See docs/usage-flow-v0.3.0.md for the complete workflow.

Current limitations

• The included vault is still an extracted real workflow, not a fully synthetic demo.
• The scripts are Windows/PowerShell-first, although the Python code uses the standard library.
• The setup skill creates a paper-writing harness. To adapt it to another domain, rename layers and update constants in the generated scripts.
• The public release omits private documents, datasets, generated full-evidence JSON snapshots, and workstation-specific paths.

v0.2.0 - Updated knowledge vault and portable local RAG

v0.2.0 - Updated knowledge vault and portable local RAG

Self-Learning Library v0.2.0 refreshes the public knowledge vault with the latest rule-layer iteration and makes the retrieval scripts easier to run outside the original workstation.

What is upgraded

• Updated the layered paper_writing_obsidian_vault through the 2026-06-26 iteration.
• Added new SCI paper-structure rules for abstract, introduction, theory, numerical simulation, discussion, and conclusion sections.
• Added mentor-derived SCI English translation and grammar rules as focused reusable notes instead of one large source note.
• Added workflow-governance rules for mandatory lightweight RAG startup, RAG repair, script reuse before creating helpers, knowledge-base maintenance ownership, and skill-promotion review.
• Refreshed supervision and no-regression notes for DOCX formula formatting, manuscript-body purity, chat-first Word insertion, and translation checks.
• Added a local fallback retrieval path in tools/kb_rag.py so the repository can run without a private total-agent-memory checkout.
• Made local paths configurable through environment variables instead of hard-coding one workstation layout.
• Redacted public export artifacts that could expose local absolute paths or full private source text.

Advantages over v0.1.0

• More complete rule coverage: v0.1.0 mainly demonstrated the harness shape; v0.2.0 includes a richer, iterated knowledge base with 96 change-log notes, 85 paper memories, 15 intermediate writing rules, 12 workflow-governance rules, 15 final generalized rules, and 15 supervision notes.
• Better translation behavior: SCI English grammar rules are now split into targeted retrieval units, making article translation and polishing easier to govern with lightweight RAG.
• Safer publishing posture: raw source-material exports and full-evidence JSON snapshots are omitted from the public release, while derived reusable rules remain available.
• Better portability: the core scripts compile and the lightweight RAG command can run in a fresh clone without private local modules.
• Stronger workflow discipline: the harness now records explicit boundaries for ordinary writing tasks, workflow-governance retrieval, no-regression checks, and script reuse.

Quick verification

The release was checked with:

py -m py_compile tools\kb_rag.py tools\paper_iteration.py
py tools\kb_rag.py --query "SCI English grammar translation article rules" --limit 3
powershell -NoProfile -ExecutionPolicy Bypass -File .\run_paper_iteration.ps1

Current limitations

• The included vault is still a real exported paper-writing workflow, not a synthetic demo vault.
• Some notes mention manuscript filenames as evidence labels, but the manuscript files themselves are not included.
• Convenience commands remain Windows/PowerShell-first.
• The local fallback RAG is intentionally simple; users with a richer memory backend can set TOTAL_MEMORY_SRC to reuse their own retrieval implementation.

Upgrade guidance

Existing users can pull the new release and rerun:

py tools\kb_rag.py --query "your task keywords" --limit 3
powershell -NoProfile -ExecutionPolicy Bypass -File .\run_paper_iteration.ps1

Optional environment variables:

$env:TOTAL_MEMORY_SRC = "C:\path\to\total-agent-memory\src"
$env:TAM_MEMORY_DB = "C:\path\to\memory.db"
$env:SCI_MEMORY_SKILL = "C:\path\to\SCI-memory\SKILL.md"
$env:SELF_LEARNING_LIBRARY_PYTHON = "C:\path\to\python.exe"

v0.1.0 - Local self-learning knowledge-base harness

v0.1.0 - Local self-learning knowledge-base harness

Self-Learning Library v0.1.0 introduces the first public shape of a local-first knowledge-base harness for AI-assisted work.

What is included

• A layered Obsidian-compatible Markdown vault.
• Codex-facing root instructions in AGENTS.md.
• A project harness workflow in PROJECT_HARNESS_WORKFLOW.md.
• Lightweight local retrieval with tools/kb_rag.py.
• Full graph-style iteration and rule promotion with tools/paper_iteration.py.
• PowerShell entrypoint run_paper_iteration.ps1.
• Example exported manuscript-assistance vault.
• Documentation for lightweight retrieval, full iteration, no-regression guards, and domain adaptation.
• Security and publishing guidance.

Who should try it

This project is useful if you want an AI assistant to:

• retrieve relevant rules before a task;
• avoid repeating past mistakes;
• separate concrete evidence from stable rules;
• detect unresolved conflicts;
• promote repeated lessons into reusable knowledge;
• keep a local, inspectable, file-backed memory structure.

Current limitations

• The included vault comes from a paper-writing workflow, although the mechanism is domain-independent.
• Convenience scripts are Windows/PowerShell-first.
• Windows users should enable Git long paths and clone into a short local path because the example vault contains descriptive long note filenames.
• The repository currently includes a real exported example vault, so users should review privacy and publishing guidance before adapting the pattern.
• Retrieval and rule promotion are intentionally simple and local-first.

Suggested repository description

Local-first self-learning knowledge-base harness for AI agents. Retrieve rules, avoid regressions, and promote repeated lessons into stable reusable knowledge.

Suggested topics

ai-agent, ai-memory, knowledge-base, self-learning, local-first, codex, rag, obsidian, markdown, workflow, research, writing, automation, python