A source-backed knowledge system for turning research, conversations, and agent work into maintained, interlinked Markdown.

Current public identity: WEIPING_WIKI / Weiping Wiki.

Historical aliases preserved for compatibility: vipin wiki, vipinknowledge, and vipin-wiki. Older paths, slugs, prompts, and automation snippets may still mention those names until they are safely migrated; treat them as aliases, not the current project identity.

The wiki is the public crystallization layer. Day-to-day agent memory and collaboration now live in agentmemory; stable facts, rules, source notes, and reusable answers are promoted back into wiki/ when they deserve a durable public home.

The old custom Agent Hub is retired. Historical pages may still describe it, but active coordination should use agentmemory.

Agents should use:

• memory_recall / smart search for active memory retrieval.
• memory_save for important decisions, lessons, configurations, and findings.
• memory_signal_send / signal reads for cross-agent handoffs.
• action/checkpoint tools when a task needs coordination state.

Markdown files under memory/ are now a historical/superseded layer unless the user explicitly asks to work with them. Reusable public-safe knowledge should be crystallized into wiki/ pages instead of requiring every agent to dual-write session files.

Agents should not wait for the user to name a skill. For non-trivial work, classify the task intent, inspect available skill metadata, read the matched SKILL.md, and follow that workflow before acting.

Primary skill roots:

Research-audit work remains strict: use the appropriate ARIS skill and preserve evidence labels, baseline boundaries, and experiment integrity.

python scripts/wiki.py status
python scripts/wiki.py health
python scripts/wiki.py catalog
python scripts/wiki.py lint
python scripts/wiki.py search "agentmemory"
python scripts/wiki.py context L1 --query "whole-computer maintenance"
python scripts/wiki.py context L2 --query "uncertainty recommendation" --graph --max-chars 12000
python scripts/wiki.py context L2 --query "uncertainty recommendation" --graph --json
python scripts/wiki.py maintain --scope whole-computer --json
python scripts/wiki.py obsidian export --json
python scripts/wiki.py obsidian commands --json
python scripts/wiki.py obsidian quick "whole-computer maintenance"
python scripts/wiki.py obsidian backlinks weiping-wiki-maintenance-system
python scripts/wiki.py crystallize --title "How to ingest sources" --type query --from-file notes.md
python scripts/wiki.py lifecycle --json
python scripts/wiki.py graph stats
python scripts/wiki.py graph neighbors llm-wiki --depth 1
python scripts/wiki.py search "llm wiki" --graph --semantic
python scripts/wiki.py scrub raw/inbox/new-source.md
python scripts/wiki.py health --fix --dry-run
powershell .\scripts\computer-inventory.ps1
powershell .\scripts\wiki-maintenance-audit.ps1
powershell .\scripts\wiki-maintain.ps1 -Scope whole-computer -Json
powershell D:\agent-resources\skills\vipin\workstation-maintenance\scripts\New-WorkstationInventory.ps1 -OutputDir ".wiki-tmp\workstation-maintenance"
powershell D:\agent-resources\skills\vipin\workstation-maintenance\scripts\New-MovePlan.ps1 -ManifestPath "<manifest.json>"
powershell D:\agent-resources\skills\vipin\workstation-maintenance\scripts\Test-MovePlanBatches.ps1 -MovePlanPath "<move-plan.json>"
powershell D:\agent-resources\skills\vipin\workstation-maintenance\scripts\New-ApprovalPacket.ps1 -MovePlanPath "<move-plan.json>" -PreflightSummaryPath "<preflight-summary.json>"
powershell D:\agent-resources\skills\vipin\workstation-maintenance\scripts\Invoke-ApprovedMoveBatch.ps1 -MovePlanPath "<move-plan.json>" -BatchId "<batch-id>" -PreflightOnly
powershell D:\agent-resources\skills\vipin\workstation-maintenance\scripts\New-DriveRootOrganizationPlan.ps1 -DriveRoot "D:\" -OutputDir ".wiki-tmp\workstation-maintenance"
powershell D:\agent-resources\skills\vipin\workstation-maintenance\scripts\Invoke-DriveRootOrganizationPlan.ps1 -PlanPath "<d-drive-root-plan.json>" -PreflightOnly

The file move-plan command defaults to a 30-day age gate and 100-item batch cap for live manifests; recent files are deferred for review instead of placed in executable batches. Test-MovePlanBatches.ps1 is a non-moving readiness check for every proposed batch, not approval to move files. D-root organization plans move eligible immediate root directories under D:\_Organized\<bucket>\_RootDirs\ and leave old paths as junctions; locked roots stay classified in place until a later retry. Once broad approval is granted, future agents should execute currently passing low-risk batches without asking for trivial per-batch confirmations.

Compatibility wrappers still exist for older workflows, but scripts/wiki.py is the canonical surface.

context L2 now returns citation-ready evidence instead of title-only previews. It validates catalog structure, page bindings, content digests, and corpus integrity; a missing, stale, corrupt, or tampered cache is rebuilt read-only in memory without mutating the repository. Catalog generation and retrieval reject Markdown paths that resolve outside the public wiki/ tree. Retrieval ranks public pages, reserves a seed-balanced minority of the result set for one-hop graph discoveries, selects the most relevant heading-scoped excerpt, and enforces both total and per-source character budgets.

Every excerpt carries a pack-local citation ([W1], [W2], ...), authoritative wiki/ path, retrieval route, section, and page SHA-256. Provenance is rebound from the live page rather than trusted from catalog metadata. Safe repository-relative source paths and public HTTP(S) locators may be emitted; absolute/UNC/drive-relative paths, traversal, URL credentials/query strings, and private-layer paths are omitted with counts. This prevents a default context pack from publishing workstation layout or private provenance merely because it exists in maintainer frontmatter.

Use --json for agent pipelines. Successful JSON mode keeps stdout machine-parseable; usage and runtime failures return a JSON error object on stderr with a non-zero exit status. The documented --root option works both before and after the command.

The implementation is dependency-free and locally adapted from four complementary public patterns; no upstream source code is copied:

• Microsoft GraphRAG local search combines graph relationships with ranked text units and filters them into a bounded context window.
• LlamaIndex CitationQueryEngine exposes citation granularity and source-node metadata.
• Quartz backlinks and graph view make linked-note neighborhoods navigable.
• Khoj search demonstrates local knowledge search over document chunks while keeping self-hosting viable.

Graph expansion is deliberately a minority of the pack: direct lexical evidence remains primary, and high-degree generic hubs are discounted. --max-chars limits the sum of excerpt text; metadata is emitted separately so downstream agents can audit what they received.

The Wiki quality GitHub Actions workflow runs on both Linux and Windows, compiles every Python script, runs the full standard-library unit suite, enforces public-wiki link/privacy gates, smoke-tests structured evidence output against the real corpus (including provenance safety and deterministic budgets), runs the secret/path safety check, and verifies that read-only retrieval never rewrites wiki/catalog.json.

• Use the .wiki-schema.md confidence taxonomy: EXTRACTED, INFERRED, AMBIGUOUS, UNVERIFIED.
• Preserve public/private boundaries. Public pages must not expose secrets, tokens, private chats, private docs, or sensitive personal data.
• Keep research project claims inside their evidence gates. Do not change experiment progress, datasets, checkpoints, or server state from this repo.
• Stage only scoped files. Existing unrelated dirty work belongs to its owner.
• Infrastructure changes must update the relevant operating docs in the same commit.
• Whole-computer maintenance, local project routing, or file-organization work should start from whole-computer project map; D-drive infrastructure detail stays in D-drive project map so agent runtime cleanup stays separate from research experiments. Physical drive organization must use the shared workstation-maintenance skill, produce a dry-run manifest or D-root organization plan first, defer recent files and cap file batches by default, optionally preflight the full plan, exact batch, or D-root plan, and move only approved batches with rollback manifests. Broad approval counts for all currently passing low-risk batches.
• Continuous WEIPING_WIKI maintenance is documented in WEIPING_WIKI maintenance system. Weekly automation should use gpt-5.5 with xhigh reasoning when supported, report first, update only curated scoped files, validate, then commit and push when real evidence changed. The historical vipinknowledge-maintenance-system slug remains an alias for older links.
• The WEIPING agentic roots are connected through WEIPING agentic project constellation: keep routes and artifact handoffs tight, but do not make repos depend on each other's private .env, cache, DB, runtime state, or generated reports.
• Obsidian-compatible local-first features are documented in Obsidian feature parity. Use python scripts/wiki.py obsidian export --json to refresh vault config, Bases, Canvas, command palette, templates, slides home, workspaces, and the dashboard.

Read AGENTS.md first. It defines the mission, source hierarchy, collaboration tone, agentmemory policy, implicit skill routing, cross-project boundaries, validation gates, and commit discipline.

Agent Hub references in older pages are historical unless explicitly marked active by a newer rule. The current operating model is agentmemory-first.