spikelab
UserThe Multiplai Claude Code plugin marketplace — a memory engine plus seven skill packs (dev, PM, research, media, messaging, writing, Apple), 44 skills, for Claude Code.
Categories
Indexed Skills (45)
buildme
Full bootstrap conductor - from idea to working code. Orchestrates interview, research, spec generation, and autonomous TDD implementation via a deterministic Python pipeline.
analyze-context-router
Analyze memory retrieval logs for routing accuracy, false negatives, token efficiency, and system health. Use when the user wants to evaluate how well the memory loader is performing, check retrieval quality, audit retrieval routing, or identify improvement opportunities. Triggers on "analyze memory retrieval", "how's the memory loader doing", "memory retrieval analysis", "check retrieval quality", "memory loader performance".
codebase-walkthrough
Generate an interactive walkthrough of any codebase. Explores the code systematically, then produces both a Markdown document and a self-contained HTML file with step-through navigation. Use when the user wants to understand a codebase, onboard onto a project, create documentation for a repo, or get a guided tour of how code works. Triggers on "walk me through", "explain this codebase", "how does this project work", "onboard me", "codebase tour", "walkthrough".
devops-gcp
Working operator's knowledge of Google Cloud Platform for DevOps tasks — Cloud Run, Cloud SQL, IAM, Terraform, Cloud Logging/Monitoring, Cloud Storage, networking, and multi-project setups. Load when the task touches GCP. For project-specific conventions, also load the matching reference file under `references/conventions/`.
e2e-test
End-to-end testing for web applications. Supports frontend (browser-based) and backend (API) testing modes. Spec-aware — reads OpenSpec artifacts when available, falls back to codebase discovery. Use when the user wants to validate that an application actually works, not just that tests pass.
learn-stack
Generate an interactive framework learning guide from any codebase. Triggers: learn the stack, learn the framework, teach me this framework, I don't know [framework], framework guide, learn-stack, what framework concepts do I need, how does [framework] work
propose-skill
Analyze session patterns and propose new skills. Use when Claude notices repeating workflows, when the skill-creation nudge fires, or when the user says "formalize this", "make this a skill", or "propose a skill". Reads session context to identify recurring tool sequences, command patterns, or file editing flows that could be automated.
skill-creator
Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations.
swift-build
Build, test, and manage iOS/macOS projects from any environment. Handles the SSH bridge when running inside a Docker container — Claude calls the script, reads structured output, never touches SSH directly.
job-application
Draft tailored resumes and cover letters, then generate PDF applications. Use when the user wants to apply for a job, create a resume, write a cover letter, generate an application PDF, or mentions a job description to apply for. Triggers on "apply for", "draft application", "job application", "resume", "cover letter", paste of a job description, or explicit /job-application invocation.
dream-remember
Review and apply pending memory updates from the learnings backlog. Checks .multiplai/dreams/ for a pre-generated Dream proposal; if none exists, generates one. Then presents updates grouped by target file, waits for approval, applies edits, and cleans up processed learnings files.
memory-health-audit
Full memory system health audit — cross-correlates retrieval logs, diary, learnings, and memory file structure to assess what's working, what's wasteful, and what to change. Broader than the /multiplai-context:health check (which focuses on infrastructure). Use when the user asks 'how's memory doing', 'audit the memory system', 'what patterns have you noticed about memory', 'memory health check', or periodically (monthly) to track system health over time.
code-review
Thorough code quality review for any codebase or code snippet. Use when the user wants to review code, evaluate code quality, audit a codebase, review a PR, assess whether code is good or bad, or get improvement recommendations. Triggers on "review this code", "is this code good", "code review", "review the codebase", "what could be improved", "audit this code", "evaluate this code".
plan
Author self-contained, executable implementation plans that carry their own completion contract — verifiable "Done means" criteria, explicit constraints/out-of-scope, and stop-and-ask gates. The resulting plan file can be handed to a fresh session ("implement the plan"), fed to a goal runner, or used as buildme input, with no conversation context needed. Triggers on "write a plan", "make a plan", "implementation plan", "plan this out", "draft a plan for", or explicit /plan invocation.
security-review
Deep security audit for any codebase or code snippet. Use when the user wants to check if code is secure, find vulnerabilities, do a security audit, assess security posture, review code they're downloading or evaluating for safety, or check for supply chain risks. Triggers on "is this secure", "security review", "security audit", "is this safe to use", "check for vulnerabilities", "find security issues", "audit security", "check this dependency", "is this library safe".
think
Critical thinking toolkit — audit conversations for assumptions, biases, unchecked facts, premature convergence, and missed framings. Three modes. Quick (default, no args) — read the conversation and deliver a concise critical audit report, non-interactive. Focused (`/think focused` or `/think about X`) — interactive critical thinking on a specific problem, 10-15 exchanges applying Skeptic/Sage/Strategist lenses. Deep (`/think deep`) — full structured decision process for big life/career/strategy decisions with phased document. Triggers on "/think", "check my thinking", "audit this conversation", "am I missing something", "what am I not seeing", "sanity check this". Complements /interviewer (which extracts hidden info) — this skill examines and stress-tests info already on the table.
excalidraw
Generate and iteratively refine Excalidraw diagrams for architecture, design exploration, and visual communication. Creates .excalidraw files that open in VS Code Excalidraw extension for drawing on top of. Triggers on "draw a diagram", "create an excalidraw", "diagram this", "visualize this architecture", or explicit /excalidraw invocation.
host-browser
Drive the user's REAL logged-in Chrome on the macOS host (via the `ab`/agent-browser bridge) instead of an ephemeral test browser. Use whenever a task needs a real, persistent, fingerprint-genuine browser session — logging into a site, filling a form, scraping a JS/bot-walled page, automating a signup, or grabbing a verification email from a temp-mail service. Provides a reliable connect procedure plus human-pacing/anti-detection helpers. Triggers on "use the browser", "drive Chrome", "log into", "sign up for", "the page is bot-blocked / 403 / JS-rendered", "grab the verification email".
transcribe
Transcribe audio files to text using mlx-whisper. Use when the user wants to transcribe audio files (.mp3, .m4a, .wav, etc.) to text, convert speech to text, or extract text from audio recordings.
youtube-transcript
Download transcripts from YouTube videos. Use when the user provides a YouTube URL and wants the transcript, captions, subtitles, or text content from the video. Supports subtitle download (fast) with audio transcription fallback.
gmail
Work with the user's Gmail. Today it can do exactly three things and nothing else: search the inbox, read one inbox message, and create a draft (it never sends, and cannot see anything outside the inbox — archive/sent/spam/all-mail are unreachable). Use when the user wants to check/search their inbox, read an email, or draft a reply/new email for them to review and send manually. Triggers on "check my inbox", "search my email", "email from X about Y", "read that email", "draft a reply", "write an email to", "gmail".
slack
Read, search, and post to Slack as the user (via their own xoxp user token — no bot, sees exactly what they see). Use when the user pastes a Slack message link (…slack.com/archives/…) and wants it read, asks what someone said in a channel or DM, wants to search Slack for a topic, or wants to post/DM a message. Triggers on "read this Slack thread", "what did X say in Slack", "check #channel", "search Slack for …", "post to #channel", "DM X on Slack", "send a Slack message", "sync my Slack", or a pasted Slack permalink.
pm-jtbd-synthesis
Synthesize Jobs-to-be-Done from one or more customer interview transcripts. Extracts Forces of Progress (push, pull, anxiety, habit) with verbatim quote + line-number attribution, clusters jobs across transcripts, and emits three artifacts — a full synthesis report, job stories in canonical When/I-want/So-I-can format, and an Opportunity-Solution Tree stub. Triggers on "synthesize calls", "synthesize interviews", "jtbd", "jobs to be done", "discovery synthesis", "what did customers say", "interview synthesis", "forces of progress", or when the user points at a folder of call transcripts and asks for product insight. Composes downstream of `transcribe` (audio → text, multiplai-media plugin) and `extract-insights` (general extraction, multiplai-research plugin); composes upstream of `pm-persona-codifier`.
pm-persona-codifier
Codify customer/user personas into canonical, source-attributed persona docs that serve as stable references for downstream product work. Takes raw archetype notes, interview transcripts, JTBD synthesis output, or stated archetype hypotheses; clusters signals into N personas; fills a 9-section template per persona with quote-level source attribution; cross-checks for overlap and contradiction; emits one file per persona plus an INDEX. Triggers on "codify personas", "persona doc", "persona definitions", "who are our customers", "customer archetypes", "user archetypes", "buyer personas", "ICP definition", or when the user has raw persona notes and wants them turned into proper artifacts. Composes downstream of `pm-jtbd-synthesis` (job clusters → persona archetypes); the JTBD output is the strongest possible input but not required.
pm-pr-faq
Draft Amazon-style Working Backwards documents — a fictional press release dated at launch plus an internal FAQ. Forces clarity on customer benefit, target persona, and tradeoffs *before* the team commits to building. Includes an adversarial FAQ generator (if the user provides fewer than 10 hard questions, the skill generates the rest as a skeptical investor / exec) and a stress-test pass. Triggers on "pr/faq", "pr faq", "6-pager", "six-pager", "working backwards", "launch narrative", "investor memo", "future-state narrative", "press release for", "fictional press release", "Amazon-style memo", or when the user wants to pressure-test a not-yet-built initiative by writing as if it had already shipped. Composes downstream of `pm-strategy-memo` (when strategy implies a launch) and `pm-persona-codifier` (the PR's fictional customer is a persona).
pm-strategy-memo
Draft leadership-grade strategy memos using Minto Pyramid (single governing thought → 3-5 supporting arguments → evidence) with a Working Backwards stress-test and a fresh-Claude reader-test before finalizing. Designed for product strategy memos, exec alignment docs, decision memos, positioning memos, and any document that must move a leadership team to a decision. Triggers on "strategy memo", "leadership memo", "exec doc", "alignment doc", "decision memo", "positioning memo", "exec memo", "memo to leadership", "draft a memo", or when the user has a brief, transcript, or thesis and wants to turn it into a memo executives would read. Composes downstream of `interviewer` (context-gather) and `extract-insights` (turn a transcript into a brief) — both from the multiplai-research plugin — and `pm-jtbd-synthesis` / `pm-persona-codifier` (when the memo needs to reference discovery evidence or personas).
deep-research
Conduct web research using a code-driven pipeline (plan, diverge, challenge, search, triage, read, reassess, synthesize). Asks clarifying questions first, searches via Brave/Tavily/Exa/Serper/You.com, follows links by default, and outputs at three detail levels (gist, structured, detailed).
extract-insights
Extract deep insights and nuances from articles, research papers, transcripts, or any long-form content. NOT summarization — decomposes content into core thesis, argument chain, key claims with evidence, rhetorical emphasis signals, and non-obvious implications. Strips filler. Surfaces what the author is driving home and the building blocks they use to get there. Triggers on "extract insights from", "what are the key insights in", "distill this", "what's this really saying", "extract the signal", or when user provides content and wants more than a summary.
interviewer
Ask great questions to uncover assumptions, learnings, and facts that wouldn't surface otherwise. Use when the user wants to be interviewed about a topic, think through a decision, clarify requirements, or surface hidden assumptions. Triggers on "interview me about", "ask me questions about", "help me think through", "probe my thinking on", or explicit /interview invocation. Works for product discovery, life decisions, technical design, problem diagnosis, and any context where deeper questioning reveals what surface conversation misses.
landing-page
Landing page copy creation and optimization. Three modes: create (full creation from scratch), audit (CRO review of existing page), iterate (generate copy variations for specific sections). Uses the interviewer skill (multiplai-research) for discovery when available, applies evidence-based conversion frameworks. Triggers on "landing page", "create a landing page", "landing page copy", "review this landing page", "optimize this landing page", "landing page audit".
writing
Content creation toolkit with 6 modes — brief (extract structure from braindump), cmd-brief (resolve commands in briefs), draft (transform brief into full draft), editor (copy edit for style and AI-tells), linkedin (LinkedIn posts), imagen (image prompt generation).
log-doctor
Analyze the multiplai runtime logs to find failures, anomalies, and degradation across subsystems (context_manager, extract_learnings, backfill, dream, session lifecycle hooks, etc.), verify root causes against source code, and produce a fix-recommendation report. Can focus on a single subsystem, and can actively exercise a functionality (probe mode) to verify its expected logs appear. Triggers on "check the logs", "why is X failing", "log doctor", "analyze multiplai logs", "what's broken in multiplai", "log health report", "verify X logs correctly", "exercise X and check the logs".
qmd-search
Manually search the user's resources knowledge base via qmd (semantic + keyword). Use when asked to 'search my resources', 'find that research about X', or when the automatic resources retrieval missed something.
refresh-catalogs
Manually regenerate catalog indexes with support for --force, --dry-run, and --only flags
setup
Onboarding — a 2-question quick setup by default; pass `full` for the complete interview. Populates memory files from starter templates.
memory-bank
Work with shared memory banks — git repositories of memory files a team or household shares. List and sync subscribed banks, report cross-bank collisions, turn dream-proposal items into a pull request on a bank, and run the adopt migration that moves personal memory into a bank and deletes the local copy. Triggers on 'memory bank', 'shared memory', 'team memory', 'subscribe to a bank', 'contribute to the bank', 'adopt into the bank', 'sync banks', 'why is this memory duplicated'.
backfill
Reconstruct learnings, diary entries, and now/ summaries from existing Claude Code session transcripts. Scans $CLAUDE_CONFIG_DIR/projects/**/*.jsonl, distills transcripts, runs diary-first extraction, and writes to the same files as the live pipeline. Default window: last 7 days; extends via --days N, --since DATE, or --all.
config-audit
Subtractive review of the active Claude/multiplai configuration on a ~60-day cadence — classifies every standing rule as still-serving, obsolete, or model-constraining, and writes a removals-first proposal to .multiplai/dreams/ for user review. Does NOT apply changes.
costs
Report API-equivalent costs for Claude Code usage — per chat, skill, subagent, project, model, day, git branch, or PR. Collects fresh data from session transcripts, then reports from the cost ledger. Triggers on 'what did this chat cost', 'how much did X cost', 'costs this month', 'token spend', 'cost report', 'what did that PR/branch cost'.
dream
Generate a processed-learnings proposal from the pending backlog and write it to .multiplai/dreams/ for review. Does NOT apply changes — run /multiplai-context:dream-remember to review and apply.
health
Memory audit — checks completeness and staleness of memory files, plugin infrastructure, and active configuration
deepen
Find deepening opportunities in a codebase — places where shallow modules can be collapsed into deep ones with cleaner seams, better tests, and clearer AI-navigability. Informed by the project's domain language (CONTEXT.md) and prior decisions (docs/adr/). Works across languages, with idiom packs for Python, Swift, TypeScript, and React. Use when the user wants to improve architecture, audit for refactoring opportunities, consolidate tightly-coupled modules, or harden a codebase before letting agents loose on it.
screen-demo
Turn a raw screen recording (.mov/.mp4) into a polished 1-3 minute landscape product demo video. Free + local — uses ffmpeg + PySceneDetect for editing and mlx_whisper on the macOS host (over the SSH bridge) for multilingual transcription. No SaaS, no API keys. The user provides a recording and a prose description ("keep it 90s, hook in first 10s, money shot at 2:30, lo-fi vibe") plus an optional music file/URL; the orchestrating Claude runs prep, authors an EDL, renders the reel. Triggers on "make a demo video", "edit this screencast", "turn this recording into a demo", "product demo from screen recording", "screen-demo skill".
fireflies
List the user's Fireflies.ai meetings (including ones shared with them) and pull full meeting transcripts, via the Fireflies GraphQL API with a bearer token — no MCP. Use when the user mentions Fireflies, wants a meeting transcript, or asks things like "list my meetings from last week", "pull the transcript of the X call", "what was said in that meeting". Triggers on "fireflies", "meeting transcript", "pull the transcript", "list my meetings".
now
Rebuild per-project now/ status snapshots from recent diary entries. Run after a backfill, or any time the injected project state looks stale.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.