Skip to content
View nikolas-sapa's full-sized avatar
💰
grinding
💰
grinding

Block or report nikolas-sapa

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
nikolas-sapa/README.md
Nikolas Sapalidis. The original two-line ASCII name over slow graphite and blue topographic dither contours. 17. Greece. Self-taught since January 2026. I build tools for AI agents and fix bugs in the tools I use. Below the text, a dithered folded wave travels slowly sideways. ns-ui. 542 components, installed as editable source. Catalog size does not establish uniform component quality. Open the ns-ui repository. Grip. Direct-CDP browser SDK and semantic snapshots. Snapshot-size benchmarks do not establish task-success or cost superiority. Open the Grip repository. SAC Capital. Paper execution, risk checks and reconciliation. No proven trading edge or continuously reliable autonomous operation; Sapa Fund belongs to this project family. Open the SAC Capital repository. Agentic Pipeline. Durable jobs and a rejected cost-router premise. The benchmark used CLI-token accounting, not metered API billing. No cost router is shipped. Open the Agentic Pipeline repository. Resumed-agent staleness. 2,631 runs studying when stale summaries overrule fresh evidence. Templated provenance helped. Model-written compaction remained unresolved. Open the research. Systems and developer tools. Entries show the overall project priority, not a quality score. 06. Toolfence. MCP metadata and tool-drift scanning. Metadata heuristics cannot establish runtime tool safety. Open the repository. 07. Sigeval. Statistical eval gates, with uncertainty. Statistical assumptions and optional stopping limit confidence guarantees; pre-PyPI. Open the repository. 08. Branch AI. Surfaced reasoning, captured as trees. Captures surfaced reasoning, not hidden cognition or exact model-state rewind; hosted viewer is read-only. Open the repository. 14. Helm. Agent control plane; isolation has gaps. Isolation backend is not bundled, budgets are not hard ceilings and some runtime writes remain unverified. Open the repository. 18. Skillswitch. Reversible coding-agent skill management. Profiles, catalog and audit are Claude-only; profile switching is not transactional. Open the repository. Applications and prototypes. Entries show the overall project priority, not a quality score. 09. Commitment Desk. Source-linked commitments and approvals. Prototype with incomplete safety acceptance; replay alone does not demonstrate model reasoning. Open the repository. 10. Basis Terminal. Tokenized-stock gaps, freshness and liquidity. Reference-price gaps do not establish executable arbitrage. Open the repository. 11. Order Evidence. Versioned evidence and invalidated approvals. The live campaign is incomplete; the executor is an attributed external reference implementation. Open the repository. 12. Taste Concierge. Culture recommendations using Qloo. Recommendation improvement is unestablished; important guest constraints remain unsupported. Open the repository. 13. Scope to Deposit. Quote approval and Sandbox payments. Local PayPal Sandbox prototype; AI-quality target failed and current model access is blocked. Open the repository. 15. NeuroLens. Multimodal ad-analysis heuristics. Scores are not validated neuroscience or advertising-performance predictions; NeuroPulse is the same project. Open the repository. 16. Release Memory. Graph-backed release evidence and review. Synthetic fixtures and a deterministic local demo do not establish model reasoning. Open the repository. 17. Clientcast. Git-based client updates and scope review. Public-by-link payloads have privacy implications; business outcomes are unproven. Open the repository. Experiments and integrations. Entries show the overall project priority, not a quality score. 19. NemoCascade. Verification-based model escalation. Documented provider examples use mocks, not an exercised live Nebius endpoint. Open the repository. 20. Curb. Experimental withdrawal-limit vault. Unaudited contract, acknowledged prior art; deployment pending in the reviewed snapshot. Open the repository. 21. x402 Bounty Radar. Bounty discovery with a paid API. Heuristics do not prove bounties are funded; documented paid-path evidence is limited. Open the repository. 22. FieldQuest. Browser-local AI for small outdoor quests. Narrow prototype with limited documented real-device and quality evidence. Open the repository. 23. Branch Obsidian. Branch trees embedded inside Obsidian. Dependent extension, not a separate flagship; no community-store release or real-host test evidence. Open the repository. 24. pre-commit-deno. Focused Deno hook integration. Contribution-sized implementation; acceptance and end-to-end compatibility remain undocumented. Open the repository. 35 merged pull requests across 14 external repositories, excluding two directory listings. November 3, 2026. Open the full merged external PR history. TESTING / MACHINE LEARNING. 20 merged PRs across 4 external repositories. CapyMOA: 10 merged external PRs. Highlight #450: fix(evaluation): load_model raises RecursionError on models holding an evaluator. Open all merged contributions to adaptive-machine-learning/CapyMOA. WebdriverIO: 7 merged external PRs. Highlight #15595: fix(@wdio/local-runner): report a non-zero exit code for signal terminated workers. Open all merged contributions to webdriverio/webdriverio. sktime: 1 merged external PR. Highlight #11307: [BUG] `BaggingClassifier`: stop reseeding the process-global numpy RNG. Open all merged contributions to sktime/sktime. river: 2 merged external PRs. Highlight #2025: Remove stale __all__ entries and add a resolution guard test. Open all merged contributions to online-ml/river. WEB / TOOLING. 4 merged PRs across 4 external repositories. Astro: 1 merged external PR. Highlight #18085: fix(@astrojs/vercel): make redirects match when trailingSlash is always/never. Open all merged contributions to withastro/astro. Traefik: 1 merged external PR. Highlight #13919: Add global default for aliasHeadersStrategy. Open all merged contributions to traefik/traefik. ESLint Stylistic: 1 merged external PR. Highlight #1283: fix(key-spacing): align parenthesized type values from the opening paren. Open all merged contributions to eslint-stylistic/eslint-stylistic. UnoCSS: 1 merged external PR. Highlight #5352: fix(core): expand scope placeholders in constructCSS output. Open all merged contributions to unocss/unocss. APPLICATIONS / AGENTS. 11 merged PRs across 6 external repositories. neatroots: 3 merged external PRs. Highlight #404: feat(checks): flag truncated model replies with finish_reason match rule (#386). Open all merged contributions to neatlogs/neatroots. Cap: 2 merged external PRs. Highlight #2673: perf(gpui): optimize screenshot alpha checks, flattening, and premultiplication. Open all merged contributions to CapSoftware/Cap. Omi: 2 merged external PRs. Highlight #22093: fix(test): use module fixture for prod cloud run jobs manifest to avoid timing guard timeout. Open all merged contributions to BasedHardware/omi. Presidio: 2 merged external PRs. Highlight #2311: Document RECOGNIZER_REGISTRY_CONF_FILE override for the analyzer server. Open all merged contributions to data-privacy-stack/presidio. Mastra: 1 merged external PR. Highlight #23791: fix(memory): respect user-pinned thread titles in Observational Memory. Open all merged contributions to mastra-ai/mastra. Orca: 1 merged external PR. Highlight #33223: fix(shortcuts): cycle floating workspace tabs on Ctrl+Tab when the floating panel is focused. Open all merged contributions to stablyai/orca. Credited reporter. Convex Auth. High severity. Patched in 0.0.95. Open advisory GHSA-c3rg-jwq9-3233. April 2025: Foodsave, solo top 5 of 100+ teams. January 2026: learned to code. February 2026: first paying website client. September 2026: four paying clients and about 2,000 euros earned. SCAT Advanced Quantitative, 2024. CTY Probability and Game Theory, June 2024; Economics and Entrepreneurship, January 2025. November 3, 2026: 3,926 contributions, 433 PR contributions, 85 repositories with PR contributions, including own repositories. 2,803 npm downloads across three tools, November 3, 2025 through November 3, 2026. Includes CI; downloads are not users. Open sources and counting notes. Visit nikolassapalidis.tech. Visit NikolasSapa on X. Read accessible full text, source links, dated counts and limitations in EVIDENCE.md.

Pinned Loading

  1. branch-ai branch-ai Public

    Reasoning canvas for AI CLIs — captures extended thinking from Claude Code, Codex, Gemini, and Droid as a navigable, forkable tree. Rewind, fork from any node, inject facts mid-thought. Local-first…

    TypeScript 3

  2. grip-browser grip-browser Public

    CDP-native browser SDK for AI agents. Median 1,998-token semantic snapshot vs 58,912 tokens of raw HTML across 8 real pages (16.0x). No Playwright or Puppeteer.

    Python 3

  3. ns-ui ns-ui Public

    A shadcn-compatible registry of 542 React components you install as plain source — no package, no lock-in. Tailwind v4, Motion, and react-three-fiber. Ships with a CLI and an MCP server for agent-d…

    TypeScript 43 4

  4. sac-capital sac-capital Public

    Multi-stage LLM equity research agent that anchors deterministic bytes32 commitments of its trading decisions on Mantle for post-hoc verification, executes paper trades on Alpaca, and enforces a fr…

    Python 3 1

  5. toolfence toolfence Public

    Security scanner for MCP servers — flags tool poisoning, prompt injection, tool drift, over-broad scope, and context-cost runaway across 12 checks. Run as npx toolfence <url>, CI-ready exit codes.

    TypeScript 2

  6. helm helm Public

    Open-source deployment and governance control plane for internal AI agents — hosts each agent, scopes its tools and domains, meters token spend against budgets, and gives IT a kill switch and audit…

    TypeScript 1