BayramAnnakov

Build the self-improvement loop for an agent your team already runs. Mines its session traces for the places a person corrected it, clusters them into staged improvement candidates, and derives a frozen test set from those episodes — where the agent proposes…

The execution layer of a Team OS. Turns a job you repeat — ticket → pull request, meeting → notes → tickets, call → CRM update, alert → triage — into a workflow an agent runs, with a human approval gate at the step that can't be taken back. It decides first…

The prototyping layer of a Team OS. Turns a feature description into a working, publishable prototype, then autonomously improves it across independent Generator-Evaluator rounds — grading it against YOUR team's own glossary, personas and decision principles,…

Run a synthetic user panel against a prototype, a flow, or a concept — personas built from YOUR team's brain, each walked step by step until they give up, with every finding split into what a real session would also have found versus what the model is…

Mine how YOUR team actually decides into a decisions.md — dated, sourced decision principles with statuses, scopes, and one-way doors — and generate a decide-like-<your-team> apprentice-loop skill: teammate commits to a recommendation first, the skill cites…

Evidence-first Team OS setup. Mines the host agent's local transcripts (Claude Code, Codex, …), MCP config, and existing repos to pre-populate a team knowledge repo interview, asks at most 5 questions, builds the minimal repo that answers the team's top…

Configure a personalized Robin (AI chief of staff) spec for any team. Reads the user's TeamOS knowledge repo, mines MCP config and installed runtimes to resolve ROBIN-SPEC's implementation-defined slots, drafts the team's soul.md and duty roster into their…

Generate a product-knowledge layer (a "brain") for a team's codebase - glossary and object model mined from real code, rot-rate tiers, a claims-manifest drift checker that fails commits when docs disagree with code, an eval harness, and a weekly sync skill.…

Diagnose a hard problem by building an evidence-graded "Why Tree" - a Goldratt Current-Reality Tree that fans many AI agents across the problem's branch-space, grades every node by the kind of evidence behind it (validating measured numbers' pipelines before…

Map elements of a stock-flow model to Donella Meadows' 12 leverage points, ranked by *promoted* impact for a stated goal and flagged for *wrong-direction* risk (the central Meadows warning that intuition usually pushes the right knob the wrong way). Triggers…

Critique a hand-drawn stock-flow / causal-loop diagram and identify the system archetype. Use when the user wants to analyze a stock-flow diagram, validate stocks/flows/loops, identify a system archetype (the eight from Pegasus/Senge - Limits to…

Take a trade-off ("improving X worsens Y") and dissolve it with TRIZ. Directs the model to act as the contradiction matrix — scanning the full set of 40 inventive principles + the 4 separation principles + ideality/resource analysis (which the LLM already…

Find the ONE binding constraint in a flow/process system using Goldratt's Theory of Constraints, classify it as a RESOURCE or (usually) a POLICY, and return the 5 Focusing Steps as concrete actions — Exploit and Subordinate before Elevate. Judges every move…

Commission an interactive stock-flow simulation web app from your AI coding agent (Claude Code or Codex). Use when the user wants to turn a hand-drawn or W2-graded stock-flow diagram into a runnable simulation with sliders. Triggers on Russian and English…

Diagnose how far along your AI Chief of Staff ("Robin") actually is — a 25-check, 5-layer maturity review across Identity, Sources, Skills, Autonomy, and Output that returns a rank (Intern → Junior Assistant → Trained CoS → Senior CoS → Hand of the King),…

Package the user's whole AI Personal OS ("Robin") into a portable, shareable plugin — a `.claude-plugin/plugin.json` manifest plus their identity files, best skills, and one scheduled loop, bundled into a folder and a `.zip` that installs on any machine with…

Set up a two-way chat channel for the user's AI assistant ("Robin") on Telegram OR Slack, end to end: create the bot/app, let Robin push messages to the user, let the user message Robin and get a reply that acts with their context and skills, and deploy it to…

Turn content into atomic Zettelkasten notes in ~/vault/atomic/ and keep ~/vault/index.md current. THREE modes by who drafts the note — FILE-IT (you wrote the insight, I only file it), DRAFT-IT (I draft one note from your scattered raw material, you edit a…

Initialize Robin's identity (CLAUDE.md, SOUL.md, user-profile.md) from evidence — your AI conversation history (Claude Code sessions, Codex sessions, or exported ChatGPT / Claude.ai). Extracts behavioral evidence (not self-description), surfaces recurring…

Complete Robin's birth certificate. Runs three short interviews to produce course-goals.md, goals-2026.md, and achievements.md — the three files /init-robin cannot infer from conversation history because they require explicit declaration, not behavioral…

Review a pull request (a diff or set of changed files) for correctness bugs and write findings to review.md. Use whenever asked to review a PR, a diff, or changed code.

REPLACE THIS — one-line trigger summary that tells Claude when to load this skill. Write it like a tool docstring, not a title. Concrete trigger phrases beat abstract descriptions. Example "Triggers when the user asks about <specific recurring task in your…

Walks a diff against a 5-point review checklist. Use when the user asks "review this diff", "what should I check in this PR", or pastes code with a question like "anything wrong with this?". Reads files via Glob/Grep/Read; never edits.

Walks a diff against a 5-point review checklist. Use when the user asks "review this diff", "what should I check in this PR", or pastes code with a question like "anything wrong with this?". Reads files via shell; never edits.

Grounded retrieval over cohort 4 memory seeds (roster, lesson designs, dossiers, transcripts). Use when answering any question about what was taught, who's in the cohort, what dossiers exist, or what was said in a past session.

Q&A flow that produces a complete colleague-target.md and starter Constitution from a 5-field dossier. Use this when a cohort 4 student says they want help defining their AI coding colleague's target task, or when "dossier" / "colleague target" / "what should…

Multi-perspective AI council - get 3-5 expert viewpoints on any product question, with debate, a groupthink check, and synthesis. Use for strategic decisions, competitive analysis, feature prioritization, positioning, pricing - any JUDGMENT question where…

Coach a product leader through producing a 90-day team-transformation plan. Diagnoses current AI maturity from existing artifacts, frames the work by persona (Enterprise / Speed / Founder), forces commitment to a trojan horse + transformation metric +…

Apply Karpathy's autoresearch loop (goal + mechanical fitness + mutable surface + keep-or-revert iteration) to ANY measurable workflow - code, content, sales, research, design, operations, not just ML or software. Use when the user asks to set up an overnight…

Autonomous prototype builder using the Generator-Evaluator loop. Builds any interactive prototype - pages, dashboards, flows, tools, visualizations - then autonomously improves it across multiple rounds via independent sub-agents. Based on Anthropic's harness…

Assess your product team's AI autonomy maturity across 6 dimensions with calibrated examples and generate a visual radar chart

Push an approval request to the bot's owner via Telegram and queue it in state.md. Use whenever a HITL gate fires — gated spend, human-platform account, ToS acceptance, irreversible/public action, cold outreach, CAPTCHA/3DS/verification handoff.

Get an independent RED-TEAM second opinion on a non-trivial decision — which bounty/competition to pursue, persist-or-pivot, whether an attempt is EV-positive. The advisor's value is its STANCE (an adversary who assumes you're wrong), not necessarily a…

Convene a multi-perspective adversarial panel for a HIGH-STAKES decision — the weekly strategy review and (mandatory) any avenue/rail pivot. Heavier and token-costlier than ask-advisor; use sparingly. Panel output is input you weigh and log, never a command.

Post the money bot's daily one-line P&L update to the group Telegram chat. Use at the end of every daily loop run, or when the user asks to "post the daily report".

Set up your own money bot from this template — interviews the owner, fills every blank in the constitution/charter/ledger, and hands back the human-only checklist. Use when the user says "adopt this bot", "set up my bot", "customize the starter", or clones…

Score inbound leads and outbound replies against the model in signals/qualification/scoring-model.md, draft the tier-appropriate reply, and record the score as a prediction that a human can later mark right or wrong. Writes leads/qualified-YYYY-MM-DD.csv. Use…

Build a lead scoring model (Fit + Intent + Timing) from what this repo already knows about who bought, marking every weight as observed, borrowed or guessed. Writes signals/qualification/scoring-model.md with the tiers, their distinct next actions, the…

Audit your sales funnel numbers against benchmarks, then answer an inbound lead using what the audit found. Finds which conversion rates don't look true, says what it compared against and where the benchmark came from, writes every finding to signals/ with a…

Find and score outbound leads against the ICP in this repo's CLAUDE.md. Writes a scored CSV to leads/ and files what it could not source to signals/targeting/. Every score carries the one source you would open to check it.

Audit your sales funnel numbers against benchmarks, then answer an inbound lead using what the audit found. Finds which conversion rates don't look true, says what it compared against and where the benchmark came from, writes every finding to signals/ with a…

Quality audit for PowerPoint decks. Catches layout bugs, design inconsistencies, accessibility issues, AI-generation tells, and silent generator failures.

Personalized onboarding for the AI Personal OS course. Runs a conversational interview to understand who you are, how you work, and what you want from the course — then creates your personal AI environment (CLAUDE.md, user-profile.md, course-goals.md,…

Turn a pile of session recordings into a short list of findings a colleague can verify in two minutes — instead of a pile of anecdotes. Question-scoped triage → watch the survivors → triangulate against events, database and source → grade every claim MEASURED…

Clean a product funnel before anyone makes a decision from it. Establishes what a person is, removes non-customers from the denominator, runs the standard leak checks, ranks losses by people rather than percentages, reconciles two independent sources, and…

Self-learning system that captures corrections during sessions and reminds users to run /reflect to update CLAUDE.md. Use when discussing learnings, corrections, or when the user mentions remembering something for future sessions.

Automate NotebookLM notebook creation from YouTube videos. Given a YouTube video URL, extract information about people featured in the video, research them online, create a new NotebookLM notebook with the video and research as sources, and generate an audio…

Interactive guide for creating Remotion videos - from strategic concept to rendered MP4. Use when the user wants to create a video, make a Remotion project, build a product demo video, generate a launch video, make a recurring content template, create a…