Recently updated open-source projects

hyperagent-orchestrator teaches Claude to use Hyperagent’s hosted MCP server as a remote control for existing agents. Its core loop lists agents, creates a thread, polls until it stops, and sends follow-up messages; recipes cover sequential chains, parallel fan-out, and user-in-the-loop refinement. The opening request must be self-contained because an agent cannot ask for clarification mid-run, and scores or costs remain visible only in the Command Center.

universal-ai-engineer organizes Claude Code work into ten modes, including preflight, analysis, project start, environment preparation, implementation, review, documentation refresh, ship check, self-upgrade, and skill setup. It plans before editing, asks once before global or system installs, avoids .env and secrets, and flags external accounts or keys the code needs but the repository lacks.

eslint-complexity-guardian locates an ESLint configuration, checks whether the complexity rule exists, and adds or updates only that rule. It shows a focused diff and explains the effect, while leaving other files, rules, dependencies, and settings untouched. It supports legacy config files, Flat Config, and eslintConfig embedded in package.json.

The guide walks a 10–50 person company through moving medical leave, vacation requests, employee records, and approvals out of chat and ad hoc spreadsheets. Employees submit from AppSheet, supervisors approve or reject through email buttons, approval history is recorded, and HR reports from Google Sheets; the first build is mostly configuration and copied formulas.

screen-vision treats the operating system’s accessibility tree as the primary source for desktop UI names, states, types, and rectangles, using OCR or icon vision only to fill gaps. Its CLI probes the host, captures a screenshot plus structured elements, and can click by element ID with dry-run as the default. It targets native, Win32, WinUI, Electron, game, or remote-desktop windows—not browser pages, which should use Playwright.

skill-smith is a meta-skill for creating Claude Code skills that are researched before design and checked before delivery. It delegates field research to market-intel, scaffolds each skill to the Skill-Repo-Spec, runs conformance, budget, and deduplication checks, and lists eval-lift wiring and self-evolve handoff as later roadmap work. The central distinction is that generated files are not considered usable until they pass a hard acceptance check.

ideation-engine reads the shape of an ideation request and routes it to methods such as affinity mapping, product framing, or provocation. It is strongest on transcripts, interviews, social posts, and research notes: the output must retain real proper nouns, concrete mechanisms, and implementable details instead of generic categories. A quality gate screens out vague ideas and helps turn raw source material into feature or experiment candidates.

seedance-storyboard converts a screenplay, scene description, ad concept, or even a one-line idea into a storyboard for Seedance or Jimeng text-to-video generation. It first fixes the global visual style, palette, mood, and continuity constraints, then lays out a timed shot sequence with framing, camera movement, subject action, and detail. The result is a shot-writing aid, not the video renderer itself.

PPT Design is a Claude skill for editable 16:9 PPTX decks built from a design system. It supports proposals, reports, lectures, and guides, with 45 layouts plus native tables and charts, and pays attention to Korean slide readability. It requires python-pptx and suitable fonts; fonts are not embedded, and rendered checking needs PowerPoint or LibreOffice.

five-step-algorithm forces a process review in the order question, delete, simplify, accelerate, automate. It prevents addition bias by making removal and simplification precede new wrappers, retries, agents, or optimizations. A Backward Check stops repeated rounds of fixes and asks whether the problematic step should exist at all. Use it before extending workflows, building agents, or carrying dead requirements into a migration.

Use this skill in Claude Code, Gemini CLI, or Codex to classify AI-system risk, map obligations, and prepare materials such as conformity assessments, Annex IV documentation, FRIAs, and post-market plans. Its guidance links work items to EU AI Act provisions and ISO/IEC 42001 and 27090 controls. It is decision support rather than legal advice or a substitute for notified-body assessment and certification.

self-evolve provides a methodology and deterministic harness for improving a skill, repository, or project over repeated cycles. The loop reflects, proposes, evaluates, judges, and accepts changes, while code—not the model—decides acceptance, rejection, rollback, and which signal source to trust. It supports program checks, anchor verification, and generative evaluation for open-ended work where ground truth is limited.

discord-history-export uses a headed Playwright browser for a user login, captures the target server details, exports visible channels and threads, and reorganizes the result into category-based HTML plus parallel JSON. It is a one-shot archive, not sync or monitoring. Automating a user account can conflict with Discord's ToS, and the skill advises changing the password afterward.

small-cap-deepdive starts from an investment theme or ticker, enumerates the relevant SEC filing universe, removes candidates with mechanical red flags such as going-concern risk or death-spiral dilution, and then ranks survivors for falsifiable due diligence. Runs keep manifests and valuation settings so results can be compared across versions, and the process forces disconfirmation rather than a one-sided story. The output is a landmine screen, not a buy list; human financial judgment remains essential.

Triage a commercial question across 15 data domains, detect available specialized sources, and guide setup when a source is missing. The skill delegates retrieval and synthesis to existing research tools, then adds citation checks, source tiers, corroboration, disconfirmation, timestamps, and explicit coverage gaps.

durable-loop adds persisted checkpoints, crash recovery, idempotent retries, append-only logs, and machine-checked completion to Claude Code’s built-in /loop. A verify_done rule requires consecutive passes before a run is considered converged, and reusable learnings can carry across runs. It is designed for long tasks that may outlive a process, while stricter dangerous-operation brakes are opt-in rather than enabled by default. The scripts use Python’s standard library.

Dual Subtitles SRT Skill combines English and Indonesian text into one subtitle track, keeping English above Indonesian while aligning cues, repairing timing, wrapping long lines, and rejecting overlaps or blank entries. Its scripts can probe media start offsets, shift active sidecars together, create exact-basename auto-load copies, and validate the four-line limit. It also renders frames from the actual movie for visual confirmation, so writing a file alone is not treated as proof.

Claude Blog Assistant acts as a conductor for the separate claude-blog skill set. In Ship mode, it moves a new post through eight gated stages covering the brief, draft, voice cleanup, SEO, structured data, analysis, repurposing, and final QA. In Score mode, it audits an existing post and ranks retrofits by impact versus effort. The required claude-blog dependency must be installed separately; humanizer is optional, while the analysis gate is mandatory in the full lifecycle.