skills

Run a high-thinking decision workflow that evaluates options through multiple independent model/provider perspectives, forces dissent, ranks the options, and returns one synthesized verdict with assumptions, evidence gaps, and an action plan. Bring your own API keys (Anthropic, OpenAI, and/or Gemini). Use when someone says "deep-decide", "mega decide", "pressure-test this decision", "evaluate options deeply", "which option should I pick", "rank these options", "stress-test this choice", or asks for a multi-model decision review. Do not use for pure factual research; use a deep-research skill for source discovery or long research reports, then feed its findings in here.

Apply Bogdan Baciu's personal editorial design system to websites, static HTML, prototypes, documents, slide decks, and financial-model artifacts. Use when the user asks for "Bogdan brand", "my brand", "personal branding", "brand this", "apply the design system", "rebrand this", "make this look like bogdanbaciu.com", or when building/reviewing bogdanbaciu.com-facing interfaces. Includes the Puddles sky-blue on warm-cream palette, Instrument Serif + Geist + JetBrains Mono typography, flat editorial layout rules, and the scoped .db-scope financial-artifact styling.

Use Dan's paid deep-research providers as one capability: Claude Managed Agent, Gemini Deep Research, Parallel.ai, Exa, and Firecrawl Research Index. Trigger when the user asks for deep research, a research harness, multi-angle research fan-out, five angles, Workflow tool unavailable, Gemini API deep research, Claude managed agent research, Parallel.ai research, or wants the same research methodology to work across repos such as dans-brain and bogdanbaciu-dot-com.

Preserve and distill long, high-value conversations into durable downstream context. Use when the user says to "capture this", "park this thread", "fossilize this", "save the transcript", "lock this down", "don't lose this", "distill this back-and-forth", "make this available downstream", or when a lengthy strategic exchange contains ephemeral insights, relationship reads, event prep, private disclosures, or future-use context that should not disappear between sessions. Also use when the user wants a light handoff of context without a full task-resumption handoff.

Generate copy-paste agent prompts and a coordinator playbook from a pre-built parallel work plan, or turn a cheap design dispute into competing "code wins" implementation variants. Use when the user has decomposed a large task into independent, conflict-free parallel tracks and needs formatted prompts for multiple coding-agent sessions; or when the user says "parallel dispatch", "parallel agents", "dispatch tracks", "multi-agent", "generate agent prompts", "code wins", "build competing implementations", or "settle this by building variants". Optionally layers in per-runner insight-lock capture (condensed into one cross-track insight memo at the end) and end-of-run pagecraft polish for HTML deliverables. Do NOT use to invent a decomposition from scratch, to author handover documents, or to launch/monitor agents. Do NOT use for single-session sub-agent delegation via the Agent tool — this is for spinning up separate coding-agent sessions.

Draft a weekly stakeholder update for any software project. Use when the user asks for a weekly update, Friday email, status report, sprint summary, or stakeholder progress report. Pulls commits from git and issue counts from the project's tracker (GitHub Issues by default; Linear or Jira when an MCP server, API token, or CLI is available), asks a short set of clarifying questions, and produces a draft markdown file in docs/weekly-updates/. Optionally renders an HTML email using a brand template if one is configured.

Research X/Twitter public conversation signals with Xquik. Use when the user asks to inspect tweets, accounts, launch feedback, competitor mentions, user pain, market chatter, or source evidence from X/Twitter and wants a bounded research brief rather than posting, monitoring, or bulk exporting.

Audit, migrate, and regression-test table systems across content-heavy websites, blogs, documentation sites, and admin surfaces. Use when Codex needs to normalize many ad hoc HTML table styles into a canonical table component, preserve legacy/admin table behavior, build a no-new-violations ratchet, add real-browser CSS regression coverage for table layout, or prepare a public-safe writeup/skill from a private table cleanup.

Generate hero images for blog posts using the blog repo's OpenAI image-generation scripts after verifying current official API docs for model names, tool syntax, size/quality options, and pricing. Use when the user asks to create, generate, regenerate, or batch produce hero/header images for blog posts in priv/content/posts. Also use to ingest manually-downloaded images from the art/ folder into the served location and wire them to the blog.

Analyze Codex or Claude Code session transcripts for failure patterns and propose agent-instruction improvements. Runs the four-stage pipeline (parse → filter → label → compile) via scripts/chat_analysis.py. Use when the user says "analyze my chats", "run chat analysis", "review my sessions", "what am I doing wrong", "chat analysis", or "/chat-analysis".

Summarize the current session's work into a structured handoff block and a saved markdown artifact that a fresh coding-agent session can execute on, including Codex, Claude Code, a VPS shell, or another local environment. Use whenever the user says "handoff", "hand off", "pass to next session", "wrap up for handoff", "session summary for next chat", "continue in new chat", or anything about transferring context to a new conversation or environment. Also use when the user is about to end a session with unfinished work. Includes the GitHub issue reference if one exists. Do NOT use for PR descriptions, commit messages, or release notes — those summarize code changes for reviewers, not session state for the next agent. Do NOT use to dispatch parallel agents — that's `parallel-dispatch`.

Interview the user one question at a time to resolve every open branch of a plan or decision before implementation. Use when the user says "grill me", "/grill-me", "interview me", "stress-test this plan", "poke holes", "push back on this", or asks for hard questions before committing to an approach. Also use proactively before plan mode when the user's request has unresolved design branches, ambiguous scope, or hidden assumptions. Do NOT use to test the user's knowledge of a subject — that's `quiz-me`. Do NOT use for code review or post-implementation critique. Adapted from Matt Pocock's grill-me skill.

Run an active-recall quiz on a topic, document, book chapter, codebase area, interview subject, meeting-prep topic, or study material the user wants to learn or retain. Use when the user says "quiz me", "test me on", "drill me on", "flashcard me", "ask me questions about", "help me study", "interview prep on", "I need to retain this", or asks to stress-test their understanding of a body of material. Ask one question at a time and adapt difficulty based on responses. Do NOT use to stress-test a plan or decision before implementation — that's `grill-me`. Do NOT use for trivia games unrelated to learning.

Audit Codex and Claude skill installation drift, autosync health, scheduler status, and reload requirements. Use when the user asks whether skills are installed, synced, stale, missing, or drifting from a source skills repo, or when they ask for a skill health report.

Coordinate a deliberation across multiple AI agents (Claude, Gemini, Codex, Cursor, Antigravity, etc.) using a shared markdown log. Teaches Claude to read the full log before responding, append entries with the locked preface format, follow the phase model (Research → Converge → Stabilize → Action-List → Approval → Execute → Closeout), enforce the approval gate, and hand off cleanly to the next named agent. Also scaffolds a new coordination session on request. Triggers on "agent coordination", "coordination session", "multi-agent", "round-robin", "files-received", or any mention of `agent-research/agent-coordination.md` or `agent-research/files-received/`. SKIP for generic single-model multi-turn work — this skill is only for sessions where the human is brokering between multiple distinct AI agents through a shared file.