>> Building Picado Labs
Open-source, verify-first developer infrastructure for autonomous AI software engineering.
| Project | The Problem | The Picado Edge | Contribute Now |
|---|---|---|---|
| build-with-ai | AI coding suffers from context drift, blank-canvas paralysis, and broken code. | Zero-API CLI, 11 production templates, and persistent .buildwithai/ memory (1.4k+ downloads • 30+ contributors). |
Issues |
| AI Model Router | Ballooning API costs, rate limits, and unpredictable model latency. | Sub-3ms Pareto Traffic Control Room UI with Thompson Sampling RL (3.7k+ downloads • 13+ contributors). | Issues |
| Terminal Agent | Autonomous coding agents make unverified changes and break repository builds. | Verify-first sandboxing (Docker/psutil), 12 core tools, self-healing, and rollback (1.4k+ downloads • 4+ contributors). | Issues |
| Agent Bench | Existing benchmarks test isolated toy prompts rather than multi-turn agents on codebases. | 12 polyglot benchmarks, local Ollama mode ($0 cost), GitHub PR ingestion, and 5D scoring (700+ downloads • 4+ contributors). | Issues |
Google PyTorch TPU - Contributor
PyTorch support for Google Cloud TPUs via XLA and StableHLO — google-pytorch/torch_tpu
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| Bincount Non-Negative Precondition Validation & Kernel Fix | Identified silent data corruption where negative inputs were clamped to 0 into bin 0 violating StableHLO ClampOp (§ClampOp: min <= max); implemented fused at::aminmax single TPU reduction dispatch without host synchronization penalty and added 5 targeted error unit tests |
#6199 (fixes #6107) |
| Addmv Self Shape & Dimension Validation | Discovered missing shape/dimension validation for self in csrc/ops/addmv/addmv_aten_kernels.cc leading to unexpected broadcast shapes or internal compiler panics; validated self rank and broadcast compatibility with comprehensive unit tests |
#6011 (fixes #6010) |
| Dot General Contraction Dimension Mismatch Disclosure | Discovered and documented missing vector length validation (lhs.size(0) == rhs.size(0)) in torch.dot, preventing unhandled StableHLO DotGeneralOp compiler errors |
#5773 |
Apache Magpie - Committer
Agent-assisted maintainership and vulnerability triage framework under the Apache Software Foundation (ASF) — apache/magpie
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| Core Security HTTP Dispatcher | Built dependency-free stdlib urllib engine with sandbox isolation and strict input sanitization for OSV.dev and CVE.org |
#1326 |
| Multi-Forge API Bridges | Engineered read-only API bridges for GitLab and Forgejo/Gitea with RFC-5988 Link-header pagination | #1369, #1441, #1469 |
| OSV.dev Vulnerability Aggregator | Built security triage adapter querying cross-ecosystem CVE/GHSA alias graphs across 30+ package registries | #1297 |
| Agent Guard Plugin Discovery | Fixed agent-guard dispatcher to discover contributed skill guards nested under plugins/*/skills/*/guards, restoring pre-execution security enforcement across plugin skills |
#1757 (closes #1756) |
| Public-Surface Security Embargo Scrub | Built zero-token deterministic Claude Code mod hook intercepting git and gh CLI calls to prevent security embargo leaks on public surfaces (CVE references, vulnerability classes) with fail-closed safeguards | #1676 (part of #1497) |
| Zero-Token Claude Code CLI Mods | Implemented zero-token /magpie-status and /magpie-skills Claude Code mod commands, setup drift check, and architecture guidelines |
#1601, #1667, #1552, #1541, #1550 |
| AI Agent Runtime Harnesses | Implemented runtime adapters and guides for GitHub Copilot CLI, Aider, Goose, Ollama/vLLM (air-gapped local LLMs), and Cursor | #1287, #1286, #1217, #1206, #1204 |
| CI Multi-Batch Test Normalization | Supported stdin - and normalized Windows path separators in ci-pytest-batches.py, resolving silent test drops on cross-platform CI pipelines |
#1767 (closes #1766) |
| Setup Doctor Probe Memory Caching | Held diagnostic doctor probe 3 and 5 outputs in memory with regression tests to prevent repeated subprocess re-probing | #1776 (closes #1775) |
| LLM Token Optimization | Reduced skill body tokens by ~20% and cut always-on system prompt frontmatter by ~50% (~350 tokens/turn saved) across contributor growth skills | #1487, #1489, #1481, #1482, #1483 |
| PR Management & Review Findings UI | Built live session feeds, review findings data model parser, and interactive UI component mods for PR management workflow | #1920, #1954, #1992 |
| Release Management Live Checklist Pane | Engineered verify-rc live checklist pane mod scaffolding and UI component for release candidate verification |
#2007 |
| Code Review & Quality Assurance | Conducted architectural and security code reviews on Discourse bridge transport security boundaries (#1755), triage staleness prompt fixes (#1753), transient retry backoff (#1752), Bitbucket approval writes (#1746), CI workspace test failure enforcement (#1807), and local archive read boundaries (#1811) | #1755, #1753, #1752, #1746, #1807, #1811, #1701, #1673 |
| First-Run Setup Guides & CLI Robustness | Added welcome onboarding guides for contributor-growth and utilities families, and fixed non-interactive TTY crashes in Vulnogram setup | #1702, #1669, #1742 |
Microsoft PyRIT - Contributor
Python Risk Identification Tool for AI Red Teaming by Microsoft AI Red Team — microsoft/PyRIT
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| GCG Attack Refactoring | Decoupled Greedy Coordinate Gradient attack into typed modules (CandidateProposer, CandidateEvaluator, ProgressiveScheduleController) with VRAM-bounded tensor cleanup |
#2671, #2700, #2720 (closes #2665) |
| Modern LLM Chat Template Injection Detection | Extended StaticPromptInjectionScorer to detect control and turn delimiter tokens across Meta Llama 3 (<|start_header_id|>, <|eot_id|>), Google Gemma (<start_of_turn>, <end_of_turn>), Microsoft Phi-3/4 (<|end|>), DeepSeek, Cohere Command R, and Mistral |
#3484, #4042 (fixes #3483, #4041) |
| XSS Executable Data URI Detection | Expanded XSSOutputScorer to detect executable and script-bearing data URIs across SVG (image/svg+xml) and JavaScript MIME types (text/javascript, application/javascript, application/x-javascript, text/ecmascript) across HTML tags, markdown links/images, and form actions with strict boundary safeguards |
#3400 (closes #3399) |
| AI & Cloud Credential Leak Detection | Added detection patterns for OpenAI, Anthropic, and Hugging Face API keys as well as modern GitHub fine-grained PATs (github_pat_) in CredentialLeakScorer |
#3286, #3236 (closes #3285, #3235) |
| Piped Shell & Azure Exfiltration Detection | Expanded ShellCommandOutputScorer to detect piped execution across modern shells (zsh, dash, PowerShell, pwsh, iex) and Azure CLI / environment exfiltration patterns |
#3308 (closes #3307) |
| Windows Canonical Path Traversal Scorer | Added canonical Windows proof-of-concept targets win.ini and IIS web.config with URL-encoded dot handling and strict file-end boundary validation in PathTraversalOutputScorer |
#3488 (fixes #3487) |
| SSTI & XXE Vulnerability Scorers | Added multi-engine RCE detection for template injection (FreeMarker, Velocity, Twig, Spring/SpEL) in SSTIOutputScorer and direct external DTD / netdoc / gopher URI detection in XXEOutputScorer |
#3541, #3486 (fixes #3540, #3485) |
| Package Hallucination Scorer | Added extraction of Ruby gems required via subpaths or with parentheses in PackageHallucinationScorer |
#3693 (fixes #3692) |
| Concurrent Scenario Isolation & Target Normalization | Rejected concurrent SDK scenario runs with atomic attempt accounting, and enforced matching normalizer templates in RoundRobinTarget |
#3774, #3768 |
| Adversarial Scorer Bypass Fixes | Extended SSRFOutputScorer to detect full 127.0.0.0/8 loopback block; eliminated comment syntax bypasses in SQLInjectionOutputScorer; and added user:password and IPv4 support to OpenRedirectOutputScorer |
#3234, #3230, #3260 (closes #3233, #3204, #3259) |
| Adversarial Converters & Text Normalization | Normalized newlines and tabs in MathObfuscationConverter, stripped punctuation before spacing in CharacterSpaceConverter, and added configurable safe parameter to UrlConverter |
#4060, #4044, #3402 |
| RFC 9110 HTTPTarget Byte-Length Fix | Stripped hop-by-hop framing headers (Transfer-Encoding & Content-Length) in HTTPTarget, preventing payload truncation on multilingual and emoji adversarial prompts |
#3008 |
| Attack Technique Factory | Extracted dynamic reflection and cached parameter discovery in MatrixAtomicAttackBuilder and technique resolution |
#2879, #2795 (closes #2869) |
| Multi-Process CI Test Isolation | Resolved MOSSBench and VLSU SQLite memory collisions in CI by enforcing isolated in-memory fixtures | #2781 (fixes #2775) |
| Image Converter CMYK Fix & Validation | Handled CMYK color mode conversion in BaseImageToImageConverter when output format is PNG, added eager n-gram size validation, and disallowed empty split separators |
#3258, #2911, #2912 |
| Frontend Type Safety | Adopted noUncheckedIndexedAccess in frontend configuration |
#3982 |
dynavec - Collaborator
Serverless hybrid vector database & agent orchestration engine on DynamoDB + S3 Vectors — codeforstartups/dynavec
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| Autonomous Agent Primitives | Built core ReActAgent and step-by-step Planner runtimes with dynamic schema generation from @tool docstrings and execution budget guards |
#319 |
| Extensible Tool Registry | Implemented thread-safe ToolRegistry with built-in connectors for stdlib HTTP, sandboxed Python execution, and FastMCP sessions |
#328 |
Real-Time Token Streaming (astream) |
Added async SSE token streaming for OpenAI and Anthropic models with sub-50ms first-token latency | #329 |
| Multi-Provider Chat Abstraction | Unified OpenAI, Anthropic, and AWS Bedrock Converse APIs under a single BaseChatModel interface |
#317 |
| Graph Engine Resilience | Added per-node RetryPolicy with exponential backoff, error fallback edges, and resumable GraphRunState checkpoints |
#338 |
| Hot-Tier Index Rebalancing (SPFresh) | Implemented incremental Mini-Batch 2-Means dynamic clustering with lock-free vector migration under continuous streaming ingest | #170 |
| Critical Graph Data Loss Fix | Fixed bug where adding edges or linking documents inadvertently wiped existing entity types and properties | #272 (fixes #271) |
| OpenAI Assistants & MCP Integration | Added OpenAI Assistants file_search tool adapter, official FastMCP server for IDEs, and SHA-256 chunk deduplication reducing embedding costs by 30–40% |
#214, #165, #120, #154 |
| HyDE & Multi-Query Expansion RFC | Authored accepted RFC combining multi-query generation and hypothetical document embeddings via Reciprocal Rank Fusion (RRF) | #215 |
| Offline Evaluation Suite & LLM Judge | Shipped retrieval benchmark runner (Recall@k, MRR, nDCG), pluggable LLM-as-a-judge for RAG faithfulness, and evaluation regression runs with CI gating | #185, #178, #351 |
| Collaborator Code Review & AWS Diagnostics | Reviewed and approved S3 Vectors client error translation (#346) and built aws doctor CLI command auditing IAM permissions, credentials, and endpoint latencies (#144) |
#346, #144 |
SQLite Graph Memory - Collaborator
Graph RAG on SQLite combining vector retrieval, wikilink graphs, and cross-encoder reranking — tonydzi/sqlite-graph-memory
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| Model Context Protocol (MCP) Server | Implemented a pure Python stdlib JSON-RPC MCP server exposing memory_recall for Claude Desktop, Cursor, and Cline |
#8 (closes #5) |
| Stdio Loop Fault Tolerance | Guarded stdio JSON-RPC loop against unhandled non-object JSON payloads (e.g., bare literals, arrays), returning structured -32600 Invalid Request responses without crashing the server process |
#32 (closes #22) |
| Source Execution & Child Process Resolution | Resolved sibling script execution and injected src/ into child PYTHONPATH so the MCP server executes directly from source checkouts without requiring package pre-installation, and propagated non-zero exit codes in --test mode |
#33 (closes #24) |
| Concurrency Scoping & Hygiene | Scoped temporary answer files per call to eliminate race conditions and prevented prompt pollution by suppressing stdout fallback on 0-hit queries | #10 (closes #9) |
| Granular Error Handling via TDD | Replaced swallowed exceptions with explicit isError: True diagnostics, distinguishing file I/O failures from empty recall results |
#14 (closes #13) |
| Doctor Module Health Fixes | Addressed four code review findings in doctor diagnostics module | #110 (closes #107) |
| Code Reviews & Release Verification | Reviewed and approved v0.3.1 release with macOS indexing fix and dynamic version assert (#27), dot-file indexing skip (#36), and relative index cwd resolution (#48) | #27, #36, #48) |
| Issue Diagnostics | Reported reproducible regressions including dot-file indexing in note ingestion and masked exit codes on missing embedding indexes | #25, #23 |
Other Organisations
Apache Airflow
Platform to programmatically author, schedule, and monitor workflows under the Apache Software Foundation (ASF) — apache/airflow
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| Task State Store Route Exception Translation | Implemented domain-to-HTTP exception translation in PATCH Task State Store route, preventing unhandled domain ValueError crashes and mapping them cleanly to HTTPException(404) with complete autospec unit tests |
#73753 |
| CLI Output Markup Sanitization | Escaped Rich console markup in airflowctl variable and pool command outputs, preventing rendering anomalies on user inputs |
#76419 |
| Task State Store Test Suite Refinement | Refined docstrings and strengthened assertion details across FastAPI Task State Store test suites | #73939 |
| SimpleTable Output Markup Sanitization | Escaped Rich markup in airflow-core SimpleTable output |
#76519 |
vLLM Semantic Router
Open, programmable decision layer and dynamic semantic router for LLMs and compute clusters under vLLM — vllm-project/semantic-router
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| Explicit UTF-8 Encoding for Config Pipeline | Enforced explicit encoding="utf-8" across CLI configuration operations and runtime loaders, eliminating UnicodeDecodeError crashes on Windows and non-UTF-8 locale systems |
#5339 (closes #5324) |
| Model Runtime Duplicate & Shadowing Prevention | Implemented validation rejecting duplicate and colliding model entries when name is omitted in model runtime | #5650 (fixes #5347) |
| AutoMix Cost-Aware Inversion Analysis | Analyzed and reported AutoMix cost-aware selection bug where preference and verification bonus invert on negative expected values | #5496 |
Microsoft Agent Governance Toolkit
Policy enforcement, zero-trust identity, execution sandboxing, and reliability engineering for autonomous AI agents covering OWASP Agentic Top 10 — microsoft/agent-governance-toolkit
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| Hypervisor Workspace Isolation Hardening | Fixed critical path traversal (..) escapes and sibling directory prefix collisions (session-1 vs session-10) in SessionScope.is_path_allowed, enforcing strict canonical boundary checks and rejecting traversal identifiers in cross-session access grants |
#4356 (fixes #4355) |
| Hypervisor Boundary Vulnerability Disclosure | Identified and documented traversal vulnerabilities in agent hypervisor session scoping with reproduction test cases | #4355 |
Microsoft Agent Framework
Framework for building multi-agent AI systems with declarative schemas and dynamic tool orchestration — microsoft/agent-framework
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| FunctionTool Tuple Serialization Bug Disclosure | Identified and reported bug in FunctionTool.parse_result and _make_dumpable failing to handle tuple return types, causing corruption of rich Content and Pydantic schema serialization |
#9488 |
Palinode
Memory substrate for AI agents and developer tools — Git-versioned, file-native, and MCP-first — phasespace-labs/palinode
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| POSIX Relative Path Normalization | Normalized to_rel_path() in path_guard to enforce POSIX forward slashes (/) across Windows runtimes, eliminating backslash leakage across MCP tool outputs and resolving associative memory search failures |
#228 (closes #214) |
| Non-Signalling Win32 Process Probe | Replaced signalling os.kill(pid, 0) with Win32 OpenProcess & WaitForSingleObject to probe worker process liveness on Windows without interrupting console process groups |
#215 (fixes #212) |
| Diagnostics Surface Hardening & AST Guard | Cleaned up bare issue references across diagnostic health modules and implemented automated AST test guard preventing private issue numbers from leaking to user surfaces | #207 (fixes #205) |
Angular DevTools
Inspection and diagnostics toolsuite for Angular components, signals, DI, and routing across browser extensions and MCP agent environments — pangular-inspector/devtools
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| MCP Agent Clear-Recordings Tool | Added clear-recordings agent tool to reset recorded HTTP calls, router navigations, and Analog server calls between reproduction runs, with action-level gating and panel event broadcast |
#365 |
| Component Tree Hierarchy Traversal | Fixed DOM traversal bug in walkAngularTree() that aborted recursion on non-component host elements, restoring missing child components nested inside standard HTML wrappers (<div>, <section>, <main>) |
#12 (closes #11) |
| Route Metadata & Redirect Inspector | Enhanced route extraction RPC scanner and Route Inspector UI to parse and display redirectTo targets, route titles, and metadata across static string and dynamic resolver declarations |
#10 (closes #9) |
Monorepo Doctor
AI agent skill that audits JS/TS monorepos for package boundaries, circular dependencies, caching, and CI test coverage — santoshyadavdev/monorepo-doctor
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| Scanner Runtime Dependency Preservation | Fixed bug where mixed default runtime imports with named type specifiers (e.g., import Button, { type ButtonProps } from '@acme/ui') were discarded as type-only, eliminating false negatives in circular dependency detection, boundary checks, and undeclared workspace dependencies |
#4 (fixes #3) |
PhilanthroPy
scikit-learn–native predictive analytics library for nonprofit fundraising, donor propensity, and revenue forecasting — PhilanthroPy-Project/PhilanthroPy
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| Transformer Timezone Validation Guard | Added unit test suite in EncounterRecencyTransformer guarding against invalid timezone names, ensuring accurate KeyError exceptions are raised instead of misleading tz-aware warnings |
#221 (closes #204) |
Layer5 Academy
Interactive learning platform for cloud native management, service meshes, and Kubernetes — layer5io/layer5-academy
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| Kind Lab Modernization | Streamlined prerequisites, fixed broken multi-line install commands, and updated official documentation references in Kubernetes labs | #278 (closes #277) |
| Podman Lab Fixes | Corrected broken static binary download URLs and updated container prerequisites | #276 (closes #275) |
| Observability Lab Overhaul | Refactored Prometheus and Grafana lab setup, replacing manual binary installs with streamlined prerequisites | #280 (issue #279) |
| Broken Link Remediation | Repaired dead hyperlinks and 404 targets pointing to the Meshery Playground | #274 (closes #273) |
TCS Labs Academy
Meshery enterprise learning labs and cloud-native service mesh tracks for TCS Labs — meshery-extensions/tcslabs-academy
| Contribution | Impact & Key Highlights | Reference |
|---|---|---|
| Service Mesh Performance Documentation Fix | Updated outdated SMP orientation links to the canonical specification authority at smp-spec.io |
#70 (fixes #69) |
🌐 All Contributions
| Contribution | Reference |
|---|---|
| All Merged PRs | View All Merged PRs |
| Issues Authored | View All Authored Issues |
X · LinkedIn · Email · Portfolio · Technical Blogs