A CEO main agent with seven roles runs every change through contract-first spec, architecture baseline + ADRs, isolated worktree fan-out, independent QA with an outside oracle, split-axis review, CI gate, release and a session handoff; ships a constitution, document system, dispatch protocol and a doctor self-check.
Turn a listening context into a segmented instrumental plan and composition prompts, without generating audio.
Review mobile diffs across privacy, disabled features, design contracts, tests and lifecycle behavior.
A CEO main agent with ten roles takes an app idea through PRD, design tokens, parallel feature builds, six-face review, device testing, store materials, budgeted candidate builds, TestFlight / Play internal upload with install receipts, separately-approved review submission and release, then store operations; Expo / React Native or web, with store-compliance, build-cost and submission gates built in.
Prepare truthful storefront copy, privacy answers, screenshot contracts and review notes as local drafts.
Audit implementation drift against architectural invariants, current facts, targets, gaps and ADRs.
Seedream images and Seedance videos through Volcano Ark's pay-as-you-go API: an estimate before every paid call, plain-language errors (balance, activation, moderation), results as phone file cards. Never uses the Coding Plan key.
Independently verify factual claims, quotations and research conclusions without rewriting the draft.
Measure decoding, duration, silence, integrated loudness and true peak without modifying audio.
Default cloud backup for mainland China: Baidu's official bdpan CLI with QR login, daily incremental uploads, auto-split above 4 GB, retries, size verification and VERDICT reports.
Verify bibliographic identity, author lists and publication details across manuscript editions.
Build a local bilingual landing-page draft with reusable design tokens and reviewable content blocks.
Compare frozen book identity against storefront drafts and actual deliverables.
Split narration into sentences and caption units while preserving every source character.
Inspect and (with approval) drive a local Chrome/Chromium over the DevTools Protocol: console and network errors, JS eval, computed styles, clean incognito screenshots, real keyboard/mouse input.
Choose Chinese typefaces by style. Audited inventory of 17 families; install 23 redistributable files with licenses and SHA256 checks on Linux, macOS and Windows. Official links for other fonts.
Mandarin TTS calling Volcengine Doubao Speech's official API directly: public voices or your own cloned voice, automatic Resource-Id routing, long-script chunking and a pronunciation lexicon. Zero-dependency stdlib CLI.
Use a coding agent as video editor and project manager: seven jobs mapped to tools (transcription, take selection, animation, B-roll, sound, FFmpeg QA) plus a feedback loop that trains a personal editing-style skill.
Models can't watch video. This entry points to Claude Eyes by moritzkremb (open source, MIT), which indexes video locally so agents can skim, search, zoom and measure it. Pointer only, no upstream code; install from the GitHub repo.
A seven-phase methodology for wrapping desktop or professional software in a CLI that agents can call reliably, with seven hard rules: call the real software, --json everywhere, non-zero exits with original errors, verified outputs, early bad-input rejection, offline by default, secrets only from environment variables. Adapted from HKUDS/CLI-Anything (Apache-2.0), ideas only.
Beginner setup guide with safe CLI and offline verification.
Beginner setup guide with safe CLI and offline verification.
Compares Volcano Ark Coding Plan, Alibaba Cloud Bailian Token Plan and Kimi Code (price, models, which read screenshots, overseas use), recommends one, then walks the owner through install, sign-up, key card, live test and switching. All three default models read screenshots in our 2026-10-16 test.
Daily, the biggest story on your beat as a ≤3-min vertical paper-collage explainer: real photos and web/X screenshots, palette cards, art fonts, smart captions. Three-pass script, media pre-check, add-media-then-switch-topic fallback, 20+ hard checks. ./setup.sh installs everything in the folder (no admin rights); bin/cns.py demo renders a sample with no keys. Voice, channel, platform are yours; YouTube upload optional, off by default.
Turn this year's featured color into light and dark palettes for websites, slides, covers or Agent J themes. Get CSS, JSON and a swatch preview with contrast checks. For inspiration. Independent of Pantone. Colors are approximate screen values.
A comment-operations workflow for Douyin, Bilibili, WeChat Channels, Xiaohongshu, YouTube and store reviews: scrape real comments, triage T0–T5, draft in your brand voice, review, send only after human approval with throttling, dedup via a ledger, and weekly insights. Drafts only by default.
Extract product issues, FAQ candidates and content ideas from real comment snapshots with traceable counts.
Audit reply drafts against real snapshots, deduplication records and human-review rules.
Classify real comment snapshots into T0–T5 and draft local replies with explicit human escalation.
Writes a handoff file with verbatim requirements, progress, decisions, paths, pitfalls and next step before context compaction, so the agent resumes without amnesia.
Competitor Funnel Teardown with traceable evidence and offline examples.
A CEO-orchestrated short-video channel pipeline with 11 roles — topic, script, captions, voice, visuals, subtitles, render, approval, multi-platform publish and analytics — gated by a deterministic QA CLI, with publishing always under human approval.
Compare immutable pre-publication predictions with comparable observed metrics and propose evidence-based changes.
Research a fresh topic or use your prompt, prepare transparent sprites and optional voice, then animate with Remotion and local sound effects. Includes a public sample and repeatable generation/validation commands. No channel or schedule.
Read-only disk inventory, safe retention, verified incremental archives, dormant daily schedules and once-only space advice.
Define scoped task briefs and evidence-based reports for asynchronous work.
Explain one concept with original doodles, mechanism arrows and an actionable checklist; includes a brand-free static example.
Export approved manuscripts locally to PDF or EPUB and inspect fonts, navigation and mixed-script layout.
Audit chapter assembly, duplicate paragraphs, unresolved placeholders and editorial evidence.
An agent-run digital press: topic → outline → parallel chapter writing → fact-check → edit → cover → pandoc PDF/EPUB → listing kit → final QA → human-approved publishing → retrospective, with an anti-slop quality gate (press_check.py). Gumroad, Amazon KDP, own site and WeChat drafts are all optional targets.
ElevenLabs TTS in 30+ languages with voice browsing by gender/use case/language, a voice-director playbook for delivery control, forced alignment and instrumental BGM generation. Zero-dependency stdlib CLI.
Draft localized Etsy titles, descriptions and tags from verified product facts, with an offline field validator.
Brief to finished explainer MP4: scene script, ElevenLabs or Volcengine narration with timestamps, auto captions, HyperFrames HTML motion, local render and QA.
Classify native impact, reuse compatible clients, and bind runtime evidence and paid-build approval to exact source inputs.
End-of-session retrospective that fixes root causes on the spot, commits every touched repo per project rules, and safely audits/cleans temporary git worktrees with a bundled script.
Prepare separate English, Simplified/Traditional Chinese, Japanese and Korean SRT tracks. Includes semantic-cut guidance, local structural checks and deterministic OpenCC conversion; bring your own alignment and translation tools.
Get your site discovered and cited with links by AI engines (ChatGPT, Perplexity, AI Overview): Bing + IndexNow, AI-crawler robots.txt, schema.org JSON-LD, citable content, llms.txt, monitoring and compliance red lines, with copy-paste templates.
Read Gmail over IMAP with an app password: filter by sender, time window and subject; extract body, links, verification codes and tracking numbers. Standard library only.
Send email from your Gmail over SMTP: text / HTML / Markdown, attachments, threaded replies, Sent-folder read-back. Preview by default; only --send actually sends; optional recipient allowlist.
Say good night before bed: your Agent quietly advances authorized work. Say good morning when you wake: hear what was done, reversible decisions, problems and anything needing you, then give today’s new tasks. Two bilingual skills with an optional morning routine.
Query Google Search Console from the command line: clicks, impressions, CTR and position by query/page, URL indexing status, sitemap submit, new-domain DNS verification. Uses your own OAuth client; credential files live in your workspace.
Turn a portrait and recording into talking-head or circular overlay footage. Default Turbo 1080p, about $3/minute; quote before spending. Phone secret cards, standard-library helper and free offline checks.
Agent workflow for code-based video authoring, editing and local rendering.
From an idea or article to a fact-checked talk script and a 16:9 animated web deck published with access control; optional segment-by-segment narration, word-level alignment, video and a separate SRT.
Harness-aware image routing: native image_gen for Codex; existing OpenAI, OpenRouter or Replicate keys elsewhere. Model discovery, reference edits and e-commerce presets. Product-ad variant plans, dated cost estimates and identity review.
Computer Use with staged editable drafts, media review, multi-track subtitles and audio, plus native export verification. 2.2 adds talking-head cleanup for self-recorded speech (fillers, retakes, false starts, long pauses, frozen frames) and a caption-card standard. Mac editor ZIP import tested; other platforms and engines require local validation. No private media or credentials.
Audit local landing pages for draft markers, bilingual coverage, resources, contrast and metadata.
Lets an agent drive headless LibreOffice to convert Word/Excel/PowerPoint/ODF/RTF to PDF, extract text and count pages, verifying every output; corrupt inputs are rejected up front, with --json and stable exit codes. `doctor` checks soffice and poppler and prints per-OS install commands without installing anything. Stdlib only, 20 real end-to-end tests.
Portable single-file SVG streetscape with deterministic day/night palettes and stroke tokens, optional character-animation composition and permanent samples.
A writing system for 'N things' listicle headlines and copy: curiosity-gap psychology, headline templates, number-choice table, e-commerce bullet-point and seeding-note examples, anti-clickbait rules; native English and Chinese versions.
Compiles scattered sources into a cited, cross-linked, searchable Markdown wiki — intake, uncited-source triage, parallel page writing, citation gate, archived batches, retrieval and JSONL export — driven by an offline stdlib CLI; no API keys needed.
Stdlib-only single-file SQLite store for small agent state, an append-only run log and idempotency markers so scheduled jobs never run twice. Never stores credentials.
Search the published episodes of the MacroAlpha (宏观阿尔法) Chinese channel: list by topic, date or keyword, read full texts, semantic search, and check whether a topic was covered — cite the source link. Paid per call with Agent J credits (list / full text 1, search / coverage 2; 2000 a month per paid seat, points convert 1:10), each answer shows the cost and balance; phase one opens free episodes only. No credential needed; requires Agent J 0.23.1+.
Bulk, marketing and transactional email via Mailgun: CSV batch sending (1000 per call), personalisation, enforced unsubscribe and postal address, idempotent ledger; plus domain authentication and compliance guide. Preview by default.
Build decks from a company template while keeping its master and layouts, edit existing decks, convert PPT/Word to PDF and images, turn PDF/Word into Markdown, and read, merge, split and fill PDFs; every job ends with a render-based QA (overflow, overlap, contrast, CJK tofu). Ships Chinese and English 'one person company' templates with 14 layouts, native charts and tables, and theme colours and fonts you can change in one place.
A gated workflow for brand landing pages: requirements, design presets, bilingual copy, build, deterministic QA, Cloudflare Pages preview, owner-approved production deploy, then SEO/GEO — with a stdlib CLI that writes a receipt per step.
Turns the data you exported through official channels (Google Takeout, Instagram, X, bookmarks, read-later, old blogs, mail, notes, photos, plus link lists and screenshots) into a local searchable index, then revives it as shopping lists, idea banks, life timelines, blog-to-podcast episodes and contact maps. Local-only, stdlib Python, never decrypts WeChat.
Create SVG text layouts and check PNG integrity, effective DPI and safe print bounds offline.
Prepare a local print-on-demand provider handoff using verified variants, print areas, shop identity and costs.
Reconcile an edited candidate against the original while protecting facts and fixed wording.
Create original artwork, check print readiness, prepare products through Printify, Printful or Gelato, and review Etsy drafts before approving a live listing.
Implement the owner's existing strategy locally: evaluate Jesse first for crypto, compare alternatives by fit, and optionally use TickerTrends company-demand data. Guidance only; no engine installation or orders.
Creator toolkit on Alibaba Cloud Model Studio (Qwen) via the bl CLI: see, generate and edit images, text/image/reference-to-video, TTS, ASR and local compose; results come back as phone file cards.
Losslessly archives settled, finished, documented or low-recurrence entries out of Claude Code auto-memory to shrink the always-loaded context.
React video production with still-frame review, reusable media/captions, strict timeline conversion and local export validation.
Independent community workflow for visual planning, asset reuse, SRT conversion and local export verification.
Review to Ad Copy with traceable evidence and offline examples.
Upload a local file to any S3-compatible store (R2, S3, OSS, COS, MinIO) and get a public or presigned download URL.
A CEO-editor plus 7 roles that turn a topic into a fact-checked spoken script, long-form article or product-seeding post — dedupe, evidence contract, research, draft, independent fact-check, polish with paragraph-level reconciliation, QA gate, delivery and performance retro — with an immutable per-piece ledger and no delivery without a recorded fact-check.
Credential-safety rules for agents plus a redacting secret scanner (staged/repo/workspace/history), .env quoting checker, opt-in pre-commit gate and incident/rotation playbook.
Eleven-dimension audit of your agent system producing a severity-ranked upgrade plan executed only after approval, with a read-only, secret-free inventory script.
Finds deterministic steps the LLM still improvises and proposes extracting them into stable CLIs with full signatures, executed only after approval.
Transcreate English into native-sounding Chinese, or minimally polish Chinese to remove AI-isms; built-in styles for product pages, social seeding notes and spoken scripts. Calls OpenRouter directly.
A cold-start subtraction pass over project docs that fixes duplication, misplacement, bloat and dead rules while protecting load-bearing content, with measured before/after size.
Draft platform-specific titles, descriptions, tags and cover copy from verified finished-content facts.
Convert .srt/.vtt subtitles into clean plain text: strips numbering, timings and tags, removes YouTube auto-caption rolling duplicates; paragraph and timestamp modes. Stdlib only.
★ Featured workflowResearch-first storytelling, configurable authors, covers and chapters, with optional SQL/vector memory, independent dedup and audience feedback.
Explore this workflow →Map every narration sentence to scenes and reviewable asset prompts.
No-code Stripe payments, Checkout and subscriptions with a safe CLI and Workers example.
Map localized cue text onto the exact source timestamps and audit coverage and readability.
Create editable SVG character scene clips from a brief and style reference, with a parameterized Remotion starter, original silent demo and frame-based QA contract.
Push text, photos and files to your phone or group via your own Telegram bot. Direct Bot API, standard library only; dry-run by default, --send to actually send.
Extract clean Markdown, metadata and schema.org product data from any web page locally; optional audio transcription via faster-whisper or an OpenAI-compatible API.
Route visual color, CSS theme and file workflows to human-facing tools, with precise hand-back instructions and a source-derived index.
Evaluate early-stage technology ventures with fact checking, evidence-based scoring, and investment memos; default to a protected R15 Noir web report.
Generate, edit, extend and download AI videos on Replicate (default Seedance 2.5): text-to-video, first/last frame and multimodal references, with a product-showcase short-video example and a production-brief guide. Stdlib-only CLI with dry-run.
Mix HyperFrames (HTML motion) and Remotion (React, data-driven) segments in one explainer: a reusable design-language file first, one storyboard, unified spec, serial rendering, ffmpeg concat, narration via explainer-video, smart-caption segmentation with punctuation-free subtitles, QA.
Break one local video into a style guide: a script extracts keyframes, contact sheets, a timestamped transcript and shot/word-rate/palette/static-frame stats; the agent writes the style guide and replication plan; a checker verifies. Needs ffmpeg and numpy; ASR via voxtype or any whisper-style local command; motion (HyperFrames/Remotion) and TTS are swappable. Bring your own video; no downloader, no keys.
Turn a chat into a shareable web report — and a template publishing protocol for automations: data JSON → strict schema validation → a web page or a mail-safe e-mail (`agentj report send` mails the owner). 15 templates (default R15 Agent J Noir; R15/R02/R14 have a mail version) × 12 themes, custom templates under the same rules (no scripts, no external loads), one-command publish to your own Cloudflare subdomain with expiry, access control, an author dashboard and WebMCP.
Audit footnote pairing, ambiguous sources and local cross-links in Markdown knowledge bases.
Export existing wiki pages, retrieval chunks and references into a new local directory.
Compile local sources into deduplicated, cited Markdown knowledge pages.
Every math, science, unit, statistics, finance and date calculation goes to Wolfram|Alpha instead of model mental math. Answers are labelled complete / partial / none, compound questions are split first, raw JSON can be attached, and the script uses whichever Wolfram API products the AppID has, telling an invalid AppID apart from a product that is not enabled. Standard library only.
Interviews you about a new agent-run project and generates its constitution, named document system and session-lifecycle skills; the methodology core of every workflow package.
Plan isolated worktrees, ownership boundaries and sequential integration for explicitly authorized parallel development.
Define a claim boundary and a source ledger before drafting factual content.
Audit spoken, article or promotional drafts against explicit format and disclosure requirements.
Turn one WeChat Channels link you chose into a creative-style report: mp4 download, shots and keyframes, palette, transcript, pacing and structure. You fetch and verify the third-party downloader and generate the CA on your own computer. PAC-only proxy with one-key restore. For personal study only.
Manage your own YouTube channel via the Data API v3 with OAuth auto-refresh: upload videos (thumbnail, playlist, scheduled publish, idempotency receipt), edit channel info and banner, check video status, create playlists. Every write previews by default and needs --execute; uploads and playlists default to private.
A configurable CEO-led YouTube production workflow with six specialist roles, motion-first video, five external caption tracks, private uploads and verified release gates. Requires owner onboarding; schedules are disabled.
Transcribe YouTube links and local media through the signed-in TurboScribe GUI with configurable CDP and local caching.