Promoted new cross-repo AGENTS.md entrypoints into the WEIPING route map for devtools, DELVTOOLS_PUBLIC, AGENT_RESOURCE, WEIPING_LAB, and WEIPING_COUNCIL.
Clarified Lab and Council artifact handoffs as explicit path/schema contracts rather than private runtime coupling.
Tightened UUPF usage: target the smallest skill or reference file, keep raw reports in ignored target work areas, and treat offline output as audit input only.
Recorded D:/DELVTOOLS_PUBLIC as the current local public-export checkout while preserving devtools-public as the GitHub and historical alias.
[2026-06-04 17:20] maintenance | agentic constellation and UUPF maintenance upgrade
Promoted the WEIPING agentic project constellation into a public-safe route map: WEIPING_WIKI, devtools, devtools-public / DELVTOOLS_PUBLIC, AGENT_RESOURCE / agent-resources, AGENTIC_SCIENCE / UUPF, WEIPING_LAB / vipin-lab, and WEIPING_COUNCIL / vipin-council.
Kept the relationship model tight but low-coupling: maintain aliases, first-read docs, artifact handoff descriptions, and validation commands, but do not copy private runtime state, generated reports, caches, DBs, .env files, or external project internals into WEIPING_WIKI.
Ran UUPF offline 108-pass upgrades as audit/planning input against the Codex weiping-wiki skill and the shared workstation-maintenance skill; accepted only curated script/doc/rule changes and kept raw UUPF reports ignored under .wiki-tmp.
Updated the active Codex maintenance automation to the current WEIPING_WIKI identity, gpt-5.5 with xhigh reasoning, Codex-only scheduled review, UUPF-offline audit rules, and scoped validation/commit gates.
Hardened workstation-maintenance D-root protection for D:/AGENT_RESOURCE, D:/AGENTIC_SCIENCE, and D:/DELVTOOLS_PUBLIC in addition to the existing D:/Research, D:/agent-resources, D:/devtools, D:/devtools-public, and D:/_Organized boundaries.
[2026-06-04 13:58] maintenance | WEIPING_WIKI rename and alias upgrade
Ran a D-drive root organization plan over 52 immediate D:/ children. The plan classified every root, protected D:/Research, D:/agent-resources, D:/devtools, D:/devtools-public, and D:/_Organized, and produced 0 unknown roots.
Completed 25 move-with-junction root moves into D:/_Organized/<bucket>/_RootDirs/...; old D:/<name> paths remain usable as NTFS junctions.
Left 5 classified roots physically at D:/ because Windows denied access or the root appeared locked. These are recorded as classified exceptions to retry only after locks or ACL issues are resolved.
Upgraded workstation-maintenance with D-root plan/invoke scripts, preflight-only support, resume/skip handling, drive-root rollback, and fixture coverage.
Sonnet/Opus review attempts failed because local Claude upstream credentials were disabled; a Codex read-only review found high-risk edge cases in partial move repair, junction target validation, rollback validation, and tampered protected-root plans, and the scripts/tests were hardened before commit.
Public wiki pages record bucket-level outcomes and safety rules only; local manifests remain ignored under .wiki-tmp/workstation-maintenance/.
[2026-06-01 16:55] maintenance | Active local roots and refresh audit helper
Added a detailed public-safe launch map for important non-research local roots such as legal-case automation, Veylora, Medora, Agent Town, Pretext, and portfolio repos.
Added an advisory maintenance audit script for stale pages, missing updated frontmatter, local-path pages, Agent Hub mentions, and missing counterpoints.
Kept no-README/scratch/archive roots at summary level and kept research experiments routed through research-project-workbench.
[2026-06-01 16:10] maintenance | Whole-computer vipin wiki maintenance map
Rebuilt the vipin-wiki skill around continuing maintenance, old-content refresh, whole-computer project mapping, and importance-based depth so future agents do not need a fresh prompt.
Added scripts/computer-inventory.ps1 as a shallow public-safe inventory helper for C:/, D:/, G:/, and user-profile junctions.
Kept research experiment code, datasets, checkpoints, logs, and progress outside this maintenance scope.
[2026-06-01 15:35] maintenance | D-drive project map and safety gates
Added d-drive-project-map as the first-stop routing page for D-drive infrastructure, public exports, private runtime state, and research isolation.
Added pre-push safety gate scripts for public-facing repositories and recorded that Agent Hub historical references should carry deprecated context.
Recorded endpoint-key rotation status privately because provider management APIs returned 401 without a logged-in dashboard session.
Source: Claude Code session, full codebase review of 143 Python files (32k LOC)
Notes: Project assessed at ~65% completion. Core gaps: mock truth default, no trained surrogate, toy baselines, no statistical rigor. Recommended direction: “Surrogate-guided Optimization with Trust-aware Fidelity Selection” for DAC/ICCAD. Recommended dropping LLM claim, demoting Memory layer. Priority roadmap defined (P0: real SPICE + trained surrogate, P1: trust gate experiment + SOTA baselines). Direction discussion pending user decision on 5 critical questions.
[2026-05-18 10:10] ingest | Agent Hub MCP server and multi-agent upgrade
.claude/skills/lidang-perspective/ (new skill, nuwa-distilled)
.claude/settings.json (Agent Teams enabled)
Source: Claude Code session, building on Codex architecture review
Notes: Renamed D:\cc to D:\devtools. All path references updated across 16+ files. Agent Hub provides 20 MCP tools for real-time collaboration. Daemon auto-starts on boot.
User report that prior cc-family collaboration felt like low-context prompting rather than real delegation.
Existing local CC workflow, research ideation policy, and collaboration tone rules.
Notes:
Added a durable rule that partner handoffs must carry the full relevant conversation state, fixed decisions, scoped artifacts, and explicit output format.
Added a durable rule that software/project/web-reference work should read the whole relevant source material deeply enough to understand the operating pattern, not a toy README skim.
Added a source note recording the user feedback so the rule has provenance.
User report that a new chat misread “叫鲸鱼执行” as a local CLI discovery task.
Existing collaboration tone and DeepSeek role rules.
Notes:
Added a durable routing rule that 叫/让/请 + 鲸鱼/DeepSeek + 执行/看/评审/总结/分析/监视 is an explicit DeepSeek delegation request.
Clarified that Codex should gather compact read-only local snapshots for filesystem tasks before handing them to DeepSeek, and must not claim DeepSeek directly accessed local files.
Clarified that DeepSeek unavailability must be stated plainly instead of silently downgrading to Codex-only, and that 监视 defaults to a one-time check unless recurring automation is requested.
[2026-05-17 23:06] query | local cc usage environments
Preserved the distinction between cc as a local command-line Claude Code entrypoint, PixelCat as the local proxy/control plane, VSCode/OpenCode/Codex as configurable local tool surfaces, and generic Claude web/app chat as a separate environment.
Avoided recording screenshot-visible account, quota, balance, or API-key details in the public wiki.
[2026-05-16 15:49] ingest | frontend frameworks public corpus
Public metadata for 告五人 Accusefive × 藤岡靛 Dean Fujioka晴れの日（晴天）
Notes:
Recorded that Vipin likes the 小柔Channel version and also likes the original version.
Preserved both title forms, 晴天 and 晴れの日（晴天）, for later rediscovery.
Added the future rule that each liked song should get a dedicated clickable song page with playable source links when available.
Added the future rule that missing task-specific tooling should be installed into the D: drive project-local temporary/cache area rather than producing a degraded deliverable.
Installed yt-dlp.exe under .wiki-tmp/tools/yt-dlp/ for media lookup and used it to identify the official 晴れの日（晴天） playable URL.
Split the ambiguous 晴天 memory into two song entries: a likely Jay Chou Japanese-cover branch and the Accusefive / Dean Fujioka branch.
Recorded that Canvas was recently hacked/unavailable and is expected to recover on 2026-05-18.
Added the practical blocker: the user cannot retrieve course syllabi from Canvas, which blocks UMN transfer-credit review because UMN needs school syllabi.
[2026-05-17 00:50] analysis | HTML PPT skill and deck workflow
Local paper-orchestra skill files and host-integration docs under skill/paper-orchestra/
User question asking whether concrete usage had been recorded
Notes:
Added the official usage pattern: prepare workspace/inputs/, invoke paper-orchestra as the orchestrator, use agent-research-aggregator only when raw logs need structuring, and reserve individual sub-skills for targeted tasks.
Local mirror under skill/content-creation-publisher/
Installed project-local skills under .codex/skills/
Notes:
Recorded the skill’s content capture, Markdown formatting, article illustration, WeChat publishing, and X/Twitter publishing modules.
Installed the aggregate content-creation-publisher skill plus 5 direct sub-skills: baoyu-url-to-markdown, baoyu-format-markdown, article-illustrator, baoyu-post-to-wechat, and baoyu-post-to-x.
Recorded concrete usage patterns, local Bun runtime notes, and the live-publishing confirmation boundary.
[2026-05-17 02:45] ingest | frontend design and chrome automation skills
Local D-drive runtime under .wiki-tmp/tools/lark-cli/v1.0.32/
Installed skills under .codex/skills/lark-* and .codex/skills/feishu-bridge/
Notes:
Installed official Lark/Feishu CLI runtime and selected official lark-* skills for Docs, Wiki, Drive, Base, Sheets, IM, and related workflows.
Created feishu-bridge as the API-first/browser-fallback router for Feishu/Lark materials and forms.
Verified binary checksum, CLI version/help, selected skill validation, and browser-fill fallback; marked real Feishu read/write tests as pending OAuth and user-approved test resources.
[2026-05-17 14:09] query | Feishu bridge OAuth and live smoke tests
User instruction that Lidang and other wiki automation outputs should be committed after validation.
Local git inspection of the previously reported Lidang automation files.
Notes:
Added a durable rule that wiki automation outputs are official maintenance results and must be validated, scoped, committed, and pushed by default.
Clarified that false dirty states with no content diff should be normalized and reported rather than committed as meaningless changes.
Checked the Lidang files that appeared modified; their object hashes matched the index and git diff showed no substantive content changes after refresh.
User instruction to install a top-tier README skill and persist future-agent rule memory.
https://github.com/github/awesome-copilot
Local installed skill under .codex/skills/readme-blueprint-generator/
Notes:
Installed and locally adapted the readme-blueprint-generator skill for high-quality README rewrites.
Rewrote the root README using the specialist workflow with stronger first-impression hierarchy and maintainer quick-start structure.
Added the durable rule that future-agent memory requests must be persisted into AGENTS.md and relevant wiki workflow pages, then validated, committed, and pushed.
[2026-05-17 16:42] query | local cc sidecar agent workflow
User instruction to let future agents call local cc directly for multi-agent coding collaboration.
Local OpenCode / Claude Code / PixelCat setup and smoke-test results from the current session.
Notes:
Added a durable rule that substantial coding work can use D:\cc\cc.cmd as a bounded sidecar agent instead of requiring the user to open Claude Code manually.
Recorded verified versions, D-drive entrypoints, PixelCat proxy status, model smoke tests, and safety boundaries without publishing auth tokens or private chat/video contents.
Clarified that Codex remains coordinator/integrator and must verify sidecar output before edits, commits, or pushes.
[2026-05-17 19:25] query | strict cc multi-agent roles
User instruction to make future agent collaboration feel more human and partner-like.
Notes:
Added a durable rule to use warmer user-facing language and frame Opus, Sonnet, DeepSeek, and Codex-created agents as collaborators rather than impersonal tools.
Recorded that Codex remains the main coordinator, Codex-created concurrent agents can be described as Codex’s 分身, and DeepSeek should default to Pro when used.
Added an upper-level research project workbench route for the three local LLM4Rec repositories.
Expanded the three project entity pages with defended claims, current gates, startup packets, module maps, file-area boundaries, git-state reminders, and future-agent entry commands.
Recorded the coverage strategy: classify git-tracked text-like files and inventory raw/data/output/log/archive artifacts by metadata only, without copying sensitive or large contents into public wiki.
Local installed skill files under .codex/skills/wechat-video-channel-publish/
Notes:
Installed a dedicated 视频号 publishing skill after confirming the existing WeChat publisher only covered公众号 article/image-text flows.
Patched the skill build command for Windows and validated dependency install, build, CLI help, and upload help without login or upload.
Created a D-drive LLM ability-hexagon anime video project with model data, storyboard, prompt packs, generated radar/cover/subtitle assets, preview MP4, and a reviewed 视频号 upload command template.
Recorded that tencent upload, including --draft, is a live-account action requiring explicit user confirmation.
[2026-05-17 22:01] query | communication assistant skill
Local installed skill under .codex/skills/communication-assistant/.
Notes:
Installed a unified communication router skill with a shared outbox and prefill-first safety model for WhatsApp, WeChat, QQ, Feishu/Lark, and email reuse.
Added adapter references that prefer QR/login-state/session-based paths and explicitly avoid password storage.
Validated pnpm install, node --check, outbox create/show/mark/cancel/list smoke tests, and confirmed the runtime state is confined to ignored .wiki-tmp/communication-assistant/.
[2026-05-17 22:57] query | nuwa prompt for lidang skill distillation
Live repair session for Microsoft VS Code black-screen behavior.
Local VS Code paths and status output, including Code.exe, argv.json, code.cmd --status, logs, and backup directories.
Notes:
Recorded the full diagnosis and repair sequence: process/path confirmation, cache and state cleanup, clean-profile tests, renderer status inspection, hardware-rendering configuration, winget repair failure, official user installer fallback, and final verification.
Documented the Cursor misclassification and restoration so future agents verify the target executable before touching VS Code-family apps.
Preserved a fast next-time runbook while avoiding public exposure of sensitive settings.json contents.
[2026-05-18 11:51] query | PixelCat CC partner outage runbook
Local checks of D:\devtools\cc.cmd, PixelCat on 127.0.0.1:8990, and the Agent Hub daemon on port 9800.
New scripts/Test-LocalCcPartner.ps1 health check output.
Notes:
Diagnosed the failure as PixelCat/ccmax upstream credential/network state rather than a Claude Code installation, prompt, or model-name problem.
Added a reusable PixelCat/CC health check that detects the upstream_credentials_disabled 502 case before future agents spend time retrying cc.
Updated Agent Hub and collaboration docs so Opus/Sonnet/Haiku calls fall back quickly and record a clear limitation when the CC family is unavailable.
Cleaned the related wiki lint gaps by indexing the collaboration-context source note and adding counterpoint/gap sections to current multi-agent pages.
Corrected stale D:\cc references in current runbooks to the live D:\devtools runtime while preserving historical log context.
Local Clash Verge, PixelCat config, network adapter, public-IP, and health-check outputs.
Notes:
Launched Clash Verge and confirmed local proxy port 127.0.0.1:7897, but no active real Clash profile was configured.
PixelCat through the local proxy still returned upstream_credentials_disabled; direct and proxied public IP were identical.
TUN forcing attempts were overwritten by Clash Verge, no virtual adapter appeared, and all temporary PixelCat/Clash config changes were restored.
Clarified that PixelCat panel address 127.0.0.1:8990 is the local API endpoint; outbound proxy/TUN routing is separate and would not replace that displayed address.
User question about changing IP without the provider’s virtual VPN application.
Prior local PixelCat, Clash Verge, and health-check results.
Notes:
Recorded that a provider VPN/TUN app is not strictly required if PixelCat can use a trusted alternative exit such as a self-owned VPS SSH dynamic proxy, trusted HTTP/SOCKS proxy, WARP-style route, hotspot, or another physical network.
Clarified the invariant that cc.cmd must keep calling PixelCat on 127.0.0.1:8990; proxy ports such as 7897 are outbound exits only.
Added verification guidance: first prove the proxy changes public IP, then configure PixelCat outbound proxy, restart if needed, and rerun scripts/Test-LocalCcPartner.ps1.
Recorded that 429 Too Many Requests is an upstream rate-limit/retry-exhaustion signal, not evidence that the local API address should be replaced.
[2026-05-18 13:31] query | CC-family fallback to Codex parallel selves
Current workspace diff for Shunyu Yao automation outputs.
Current workspace diff for OpenAI Cookbook mirror automation outputs.
raw/lidang-public/inbox/2026-05-18-light.json.
.claude/skills/mattpocock-skills.
scripts/Test-LocalCcPartner.ps1.
Notes:
Audited the Shunyu Yao automation changes as conservative maintenance: failed crawls now preserve previous manifest entries and record crawl errors without inventing new facts.
Audited the OpenAI Cookbook mirror refresh as source-preserving maintenance: 235 entries remain, no URLs were added or removed, and the page count matches the manifest.
Audited the new Lidang light capture as public-safe failed-crawl evidence with one metadata-only X profile probe and no private-path leak.
Audited the Matt Pocock Claude skill install as MIT-licensed, provenance-matched, and secret-scan clean; nested .git metadata was moved to ignored .wiki-tmp backup so the skill can be committed as normal source files.
Confirmed CC-family partners are still blocked by PixelCat upstream_credentials_disabled, so this review used Codex-side verification rather than a real Opus/Sonnet pass.
[2026-05-18 20:25] bootstrap | OpenCode promoted to full collaboration partner
Pages created or updated:
AGENTS.md — added OpenCode to CC partner table, role assignments, collaboration patterns, and new ## OpenCode Partner Policy section
CLAUDE.md — updated to 6-agent system, added OpenCode identity and invocation path
README.md — updated to 6-agent system, added OpenCode to agent roles table and getting started
User request to retire Agent Hub, make agentmemory active, improve implicit skill use, and prepare agent-resources / devtools for public-safe open source.
Current local docs and configs under vipin-wiki, D:\agent-resources, and D:\devtools.
Notes:
Marked Agent Hub as deprecated historical infrastructure.
Recorded agentmemory as the active collaboration/memory layer and markdown memory/ as historical/superseded.
Added durable implicit skill-routing rules so agents invoke skills by task intent rather than waiting for explicit skill names.
Replaced remaining active ARIS Claude skill handoff wording with agentmemory signals/actions and explicit context packs.
Removed tracked memory/sessions/*.md dumps from the public Git index while leaving local files in place.
Removed active Agent Hub state dependencies from the wiki dashboard/token scripts; they now use agentmemory/default local token state and only read legacy metrics from an explicit environment variable.
Removed a secret-bearing key memory file from public Git, rewrote vipin-wiki history, force-pushed the clean history, and verified a fresh remote clone has zero exact-key/history hits.
[2026-06-01 16:42] maintenance | vipinknowledge continuous maintenance system
Added python scripts/wiki.py maintain --scope whole-computer --json as the canonical ignored-report command for weekly maintenance.
Rebuilt the Codex vipin-wiki skill as a mode router with separate references for scan depth, weekly maintenance, self-upgrade, and automation safety.
Added a Claude/OpenCode-visible adapter skill so non-Codex partners can trigger the same maintenance workflow.
Recorded scoped automation gates: report first, update curated wiki/skill/docs only when evidence changed, validate, then stage and push only scoped files.
[2026-06-01 16:44] ingest | University of Minnesota AZ hold guidance
User-provided pasted text about AZ holds, English proficiency scores, ESL course registration, MN Battery Test, transfer exemptions, and MELP contact information.
Notes:
Preserved the provided passage verbatim in both the raw inbox text file and the source note’s verbatim text block.
Marked the source as user-provided and not independently verified.
[2026-06-01 17:10] maintenance | commit discipline for cross-chat wiki outputs
User instruction that obvious same-session/cross-chat wiki, raw, memory, and documentation outputs should be committed rather than left behind as trivial adjacent dirty work.
Notes:
Added the durable rule that deliberate, public-safe, validated wiki/raw/memory/doc outputs should be included in a scoped commit even when they were produced by another active chat.
Preserved the guard against committing source-unclear, sensitive, conflicting, incomplete, external-project, or validation-failing work.
Obsidian official Help pages for core plugins, Bases, Properties, Web Clipper, and CLI.
JSON Canvas 1.0 specification.
Obsidian GitHub organization pages for open components such as JSON Canvas and Web Clipper.
Notes:
Recorded that Obsidian core is proprietary, so this project adapts documented behavior and open file formats instead of claiming source-level parity.
Added real local-first artifacts for opening the repository as an Obsidian vault: settings, Bases, Canvas, bookmarks, workspaces, generated dashboard, command page, slides home, daily/web-clip templates, and CLI reports for backlinks, outgoing links, search, quick switcher, outline, preview, footnotes, tags, properties, tasks, word count, random notes, external links, format audit, slides, and daily/unique notes.
[2026-06-01 17:55] infrastructure | devtools agentmemory cleanup and mem0 mapping
Reaffirmed D:/Research as a hard no-move boundary for workstation organization.
Generated a dry-run workstation inventory with 3,026 classified entries and 669 low-risk move-eligible file candidates; D:/Research entries were 0.
Tightened move-plan generation to default to a 30-day age gate and 100-item batch cap: 531 old low-risk candidates remain executable, 138 recent candidates are deferred, and executable reparse points, directories, git worktree items, and recently modified files are all 0.
Generated 13 granular approval-candidate batches by file type and part number; no physical file move was executed.
Added and tested non-moving batch preflight support; batch-downloads-installers-old preflight passed with 1 checked item and 0 moves.
Actual C/D file moves remain gated on explicit user approval of generated batch IDs.
[2026-06-01 19:56] infrastructure | workstation move-plan full preflight
Fresh local inventory workstation-inventory-20260601-201417.json and move plan workstation-move-plan-20260601-201438.json under ignored .wiki-tmp/workstation-maintenance/.
Fresh full preflight summary workstation-preflight-summary-20260601-201503.json.
Applied summary workstation-applied-summary-20260601-201604.json and per-batch rollback manifests.
Post-move inventory workstation-inventory-20260601-201722.json and post-move plan workstation-move-plan-20260601-201812.json.
Notes:
User granted broad approval for the current low-risk organization work and asked not to stop on trivial approvals.
Executed all 13 preflight-passing low-risk batches, moving 531 old files, about 307.08 MB, from scattered/download locations into D:/_Organized/Downloads and D:/_Organized/Media.
Verification found 531 original sources absent, 531 moved destinations present, and 0 applied paths under D:/Research.
Post-move inventory shows 138 remaining move-eligible candidates, all deferred by the 30-day age gate; post-move move plan has 0 executable batches.
Supplemental MinimumAgeDays 0 move plan workstation-move-plan-20260601-202730.json.
Final inventory workstation-inventory-20260601-203318.json and final move plan workstation-move-plan-20260601-203338.json.
Updated workstation-maintenance duplicate-destination handling in D:/agent-resources.
Notes:
Executed the broadly approved supplemental pass for recent-but-move-eligible low-risk candidates.
Moved another 138 files, about 490.31 MB, into D:/_Organized.
A duplicate temp-log target surfaced during execution; the skill now generates unique destination paths and rejects duplicate batch destinations before moving.
Final inventory reports 0 remaining move-eligible candidates, 0 executable batches, 0 deferred candidates, and 0 D:/Research hits.
Total approved organization moved 669 files, about 797.39 MB. No deletion was performed.