project-init
🔌PluginMoxyWolfLLC/moxywolf-plugins
|Cowork project lifecycle for MoxyWolf projects. /init-project scaffolds tailored Project Instructions. /session-start auto-resolves the project — explicit argument, then launch directory, then the most recently active project, never a picker — mounts the three standard roots, reads the previous session's handoff (cross-checked against current team-shared memory so stale procedural reminders get flagged STALE rather than silently propagated), reads Taskade/_Shared Files/_shared-memory/INDEX.md and surfaces it as a 'Team-shared rules in effect' briefing section, and probes the team-shared OpenRouter key (DR-010). /session-end asks who authored the handoff and whether it's a baton-pass, writes the canonical handoff with `author` / `for` frontmatter, requires every general behavioral reminder in the procedural-reminders section to cite a currently-existing team-shared memory file (or be tagged session-specific), supports --author / --handoff-to / --solo, refreshes each writable repo's README.md, then runs /obsidian-update. /init-project saves the full instructions on disk as the single source of truth and pastes only a thin loader stub (mounts + file-write-path override + read-the-file-first imperative) into Cowork settings, so the embedded copy never drifts out of sync with the on-disk file (DR-010). The stub skeleton and on-disk reconciliation directives live in a vault spec that both /init-project and /refresh-project-instructions read at run time, so changing a project-instruction rule is a vault edit plus a per-project /refresh-project-instructions run, never a plugin update. v0.22.0: folder access is fully mode-agnostic across online (cloud) and offline (on-computer) — each capability is detected by tool presence with a filesystem-probe fallback (detect via get_device_info or by listing the path; mount via request_cowork_directory when present, else prompt Add folder; pick via native picker when present, else device_list_dir + AskUserQuestion chips), so nothing hard-depends on request_cowork_directory and the skills run identically online or offline. v0.23.0: task-board reads moved from the retired vault kanban (KANBAN_VIEW.md) to the team Jira board (project MOXY) — /session-start Step 4b queries MOXY via the Atlassian MCP, strictly filtered fail-closed to the project's Jira label (project-<slug>); the 'Kanban project tag(s)' field now denotes a Jira label.
Part of
MoxyWolfLLC/moxywolf-plugins
Installation
/plugin marketplace add MoxyWolfLLC/moxywolf-plugins/plugin install project-init@moxywolf-pluginsMore from this repository10
End-to-end academic content pipeline: BibTeX theme mapping, perspective and voice capture, Academia.edu formatting, structure planning, iterative drafting, bibliography generation, and a 10-phase professor critique. Voice-first — voice and formatting are fixed before drafting, not retrofitted.
MoxyWolf analytics plugin — read-only reporting across data sources, one command per source. Ships the google-analytics command (GA4 Data API via a service account) with generic traffic reports plus Lens-Test campaign reports: acquisition by channel, first-touch signup source, custom-dimension segment (e.g. RPO vs non-RPO), and a one-shot markdown scoreboard with week-over-week for the Friday read. Built to grow — Microsoft Clarity, Ahrefs, PostHog, and others can be added as their own commands. Credentials come from the environment, never hardcoded; GA4 covers the acquisition end while the conversion verdict stays in PostHog. v0.1.0.
Software development execution engine adapted from Garry Tan's gstack (MIT). Code review, adversarial just-committed review via /gstack-codex-review (real Codex when the codex CLI is present, Claude fallback otherwise), security audit, debugging, QA testing, and shipping — /gstack-ship validates in a disposable worktree with a safe-fix/escalate contract (concept-ported from no-mistakes, MIT). Pairs with product-orchestrator. Browser ops via Claude in Chrome.
Energy-aware personal operations system with Inbox Intelligence and Fitness Coach. Combines Apple HealthKit/Oura (via Health Auto Export), Google Calendar, Gmail, Google Drive tasks, and an Apple-Health-aware daily workout coach.
Stage-sharded /email-* commands for the product email lifecycle: /email-activation, /email-nurture, /email-convert, /email-retain, and /email-lifecycle (the whole arc). Each runs a shared engine skill scoped to its stage and returns five things: best practices (installed authoring skills + live external research), tooling (installed skills to use + catalog adoption candidates queried from the Workforce Automation Supabase catalog, de-duped vs our installed set), a benchmark of a named product's current setup graded against best practice, prioritized recommendations aligned to the MoxyWolf reference runtime (PostHog Workflows + Resend + first-party event capture, no n8n, single body store per class), and an auto-drafted email sequence for the stage (triggers, delays, exit conditions, subjects, copy in MoxyWolf voice). Email-specific, stage-sharded sibling of marcom-audit.
Jira (project MOXY) is the single canonical task board. Sweeps non-chat sources (Google Drive Active Tasks, Google Calendar, Gmail) for genuinely-new action items, dedups them against the board (open + recently-Done), and files them to MOXY with best-practice issue authoring (Context/Acceptance/Notes descriptions, sub-items as real Subtasks, dependencies as Blocks links, Backlog default), then maintains board hygiene (priorities, column limits, In Review reviewer pairings). v0.6.0 retired the vault kanban and all Slack task flows (no Obsidian source-of-truth, no dual-authority sync-back, no Slack scraping, no #general digest) — team members file their own tickets directly in Jira, including from Slack via the Jira Slack app. Bulk writes to the shared tracker stay human-gated.
Multi-model AI deliberation with vault-aware institutional memory — structured consensus from multiple LLMs with self-learning routing, self-improving prompts, and bidirectional Obsidian vault integration.
Tooling for The Frontier Founder site. /blog-post converts any draft markdown file into a publication-ready, SEO/AEO-structured blog post — frontmatter with a meta-description excerpt, AEO structure (At a Glance, Key Takeaways, FAQ), a single @graph JSON-LD block (BlogPosting + Person + Organization + FAQPage) on the thefrontierfounder.com canonical, MoxyWolf typographic rules, a brand-aligned hero image, and slug-tied file naming, saved into the FrontierFounder repo. AEO thresholds defer to the canonical aeo-checklist in 4d-blog-engine.
Six vault-* workflow skills vendored from az9713/claude-code-obsidian. Companion to obsidian-skills (kepano's foundational five) — those handle the file formats, these run the daily vault workflows: vault-capture (quick-drop into _inbox/), vault-save (structured save to the right folder), vault-journal (Daily Journal/YYYY-MM-DD.md authoring), vault-synthesize (multi-note synthesis with citations), vault-moc (Map-of-Content authoring), vault-health (orphan / broken-link / stale-tag audits). v1.0.0.
Operationalizes the 4D AI Fluency Framework (Delegation, Description, Discernment, Diligence) for content production. The single front door for every MoxyWolf blog property: a targets/ registry (FrontierFounder wired; STIGViewer, moxywolf-website, prfaq register-only) drives canonical URL, JSON-LD, frontmatter, and hero per destination, and every post picks a publishing target + a pillar (new or existing) before drafting — the hub-and-spoke methodology, with /blog-pillar to manage pillars and their linking maps. /blog-init is writer-friendly: two folder picks, author name, hero vibe, optional live URL. /blog-voice runs a self-contained 8-question voice interview and writes a per-writer voice profile; multi-voice supported. /blog-start mounts the two directories and surfaces in-progress and unpublished pieces. /blog-pipeline runs the 4-phase pipeline: 30-day discourse sweep across reddit/X/HN/Substack/Facebook/Quora/podcasts/academic with a /council:deliberate synthesis pass, verified bibliography via bibtex-builder, draft in the writer's voice via research-pipeline/content-writer, two-tier anti-AI-slop pass, then a nonce-bound Release Owner Gate (100-point rubric). On sign-off, the pipeline ends — social derivatives are opt-in. /blog-social derives any subset of: LinkedIn Post + first-comment (feed Post 1,300-2,500 chars sweet spot, 2,900 hard cap under LinkedIn's 3,000-char ceiling, zero body links; companion first-comment file with the fixed "If you want to do your own research…" opener, the blog URL, and the 2-3 sources the Post quotes inline), LinkedIn Article (optional long-form, separate selection), Twitter/X (5-10 post thread with ≤280-char per-post hard cap, script-enforced), Facebook (single 300-500 char post with body-inline link). One voice profile, per-platform register shifts (Twitter compressed; Facebook warmer; LinkedIn professional; first-comment voice-off and utilitarian). /blog-publish ships a signed post — the writer never types a git command. Implements the Release Owner Gate from MoxyWolf's Beyond the Prompt whitepaper.