Sibyl — Deep Research MCP changelog

https://yichenwu.dev/llms.txt

− - Argus: a local testing tool for AI coding assistants, with exploratory browser−   testing, symptom rechecks, evidence reports, and test coverage tracking.−   Version 0.12.0 exports standalone Playwright tests for reproduced fresh-load−   text checks; unsupported steps need manual completion. Saved text and HTTP−   checks can also run again without a model. Claude Code and Codex plugins−   provide a guided testing workflow.− - AgentDetox: helps prevent poisoned instructions in AI agent memory and rules,−   detect suspicious changes, and clean up affected files by restoring trusted−   versions. Includes write checks, recorded change origins, and quarantine.−   Version 0.2.0 is published on PyPI as a preview. Claude Code is supported;−   Codex and Cursor hooks are experimental. Detection is best-effort, not a−   complete security boundary.+ Legal name: Yichen Wu. Preferred name: Adrien (also written Adrien Wu).+ - Argus: an open-source MCP tool that helps compatible coding agents explore+   user journeys, check state changes, and test edge cases in web apps. It tracks+   tested and unfinished goals and records reproducible bugs with evidence.+   Current release: 0.13.1. Since 0.12.0 it exports standalone Playwright tests+   for reproduced fresh-load text checks; unsupported steps need manual+   completion. Live verification and exported tests apply the same rules, and+   checks that cannot show the symptom (for example, behind a login page) are+   reported as inconclusive rather than fixed. Saved text and HTTP checks can+   also run again without a model. Setup is one command; the first session+   downloads Chromium. Claude Code and Codex plugins provide a guided testing+   workflow.+ - AgentDetox: a local Claude Code plugin that helps prevent poisoned writes,+   detect suspicious instruction changes, and clean up selected content with+   previews and backups. Preview version 0.6.0 is published on PyPI and adds an+   optional, off-by-default model check of added text. Its interactive project+   page illustrates prevention, detection, and cleanup; it is not a live agent+   session. Known-risk rules and the model check do not catch every attack.

https://yichenwu.dev/llms.txt

− - Argus: a local QA runtime for coding agents, with goal coverage, structured−   symptom checks, and evidence reports. Automatic regression uses built-in−   text and HTTP checks. The 0.7 release validation covers Chromium; live native−   macOS UI and a new paid model benchmark were not exercised.− - Sibyl: Deep Research MCP for multi-round investigations, evidence review,−   persistent jobs, and cited reports. The 0.4.0b1 beta focuses on the 12-tool−   keyless host-assisted profile; autonomous research is experimental. Live−   retrieval still misses readiness and stability targets. The beta is published−   on PyPI; install sibyl-research==0.4.0b1 explicitly. Version 0.3.0 remains−   the older stable package.−   Project: https://yichenwu.dev/projects/sibyl−   Source: https://github.com/chriswu727/sibyl+ - Argus: a local testing tool for AI coding assistants, with exploratory browser+   testing, symptom rechecks, evidence reports, and test coverage tracking.+   Version 0.12.0 exports standalone Playwright tests for reproduced fresh-load+   text checks; unsupported steps need manual completion. Saved text and HTTP+   checks can also run again without a model. Claude Code and Codex plugins+   provide a guided testing workflow.+ - AgentDetox: helps prevent poisoned instructions in AI agent memory and rules,+   detect suspicious changes, and clean up affected files by restoring trusted+   versions. Includes write checks, recorded change origins, and quarantine.+   Version 0.2.0 is published on PyPI as a preview. Claude Code is supported;+   Codex and Cursor hooks are experimental. Detection is best-effort, not a+   complete security boundary.+   Project: https://yichenwu.dev/projects/agentdetox+   Source: https://github.com/chriswu727/agentdetox+ - Sibyl: a paused research MCP project for persistent evidence, counterevidence,+   and gap tracking, with optional host-authored reports. Two comparative runs+   did not establish a clear benefit over built-in search. Current source is+   host-only; published beta 0.4.0b1 and stable 0.3.0 contain older architectures.+   Project: https://yichenwu.dev/projects/sibyl+   Source: https://github.com/chriswu727/sibyl

https://raw.githubusercontent.com/chriswu727/sibyl/HEAD/AGENTS.md

− Sibyl is a local deep-research MCP server. It turns bounded public-web− investigations into reports with inspectable evidence and explicit limitations.− ## Installation− Use Python 3.11 or newer. Report/finance extras require Pillow 13.0.0 or newer;− do not restore Python 3.10 compatibility by weakening the security floor.− This guide targets the published 0.4.0b1 beta; PyPI 0.3.0 predates these APIs.− Use the exact-version installation in README.md. The core installation needs− no search or model key. Install `sibyl-research[report]==0.4.0b1` for autonomous model-backed research− and PDF export; Markdown/JSON host reports need no optional model dependency.− 2. Call `continue_research(job_id, query, question_id)` with focused questions. Follow leads,−    investigate contradictions, and look for original sources.−    passage IDs, a conclusion, and a counterevidence note. Investigate counterevidence− 5. Submit `complete_research(job_id, revision, summary, key_findings, analysis,−    limitations)`. Summary and each finding must cite selected `[Source N]`.− Legacy checkpoints retain their previous completion behavior. The autonomous− pipeline does not populate the new host review ledger.− targets, not semantic entailment. Do not describe host reports as independently− verified. Citation numbering can change after retrieval: always use the current− ## Experimental autonomous research− `start_research(question, mode="autonomous")` uses the operator-configured model− and can incur API charges. It does not inherit a host subscription or invoke− hidden MCP sampling. Poll `get_research` for progress and the eventual report.− `continue_research(job_id)` explicitly resumes interrupted work and may incur− additional charges. Existing evidence and consumed budgets are preserved;− unfinished synthesis may be regenerated.− The compatible blocking `research(query, depth=2)` tool uses the same source− pipeline but retains only the last report in memory. Prefer persistent jobs for− long investigations.− - Model source-support checks are fallible; preserve inline unverified markers.− ## Evidence primitives− - `gather_bundle(query)`: SourceBundle 1.6 with source/passages and diagnostics.− - `gather_sources(query)`: readable rendering of that same bundle.− - `gather_evidence(question)`: host-planned atomic loop, four-step ceiling,−   ten-minute TTL. Follow its explicit next action.− - `quick_search(query)`: search hits without full-page extraction.− - `read_url(url)`: text from one public HTTP(S) page.− Use a primitive for a focused fact lookup; use persistent jobs for reports.− These APIs remain available alongside the research workflow.− Checkpoints are private local files below `SIBYL_OUTPUT_ROOT/.research`.− One runtime owns a research directory at a time. The default job allows eight+ Sibyl is a local deep-research MCP tool for AI agents. It maintains bounded+ public-web investigations, evidence-linked conclusions, and explicit gaps.+ Human-readable reports are optional outputs of the same research state.+ ## Installation and interface+ This checkout is an unreleased host-only source preview. Development paused on+ 2026-10-28 after the paired host evaluation did not pass, and no stable release+ is planned. Published 0.4.0b1 and 0.3.0+ still contain the old architecture. Use Python 3.11+ and a fresh environment+ with `python -m pip install -e .`. `sibyl-mcp` exposes eight tools with compact+ MCP `structuredContent`, mirrored as JSON text. The former keyless compatibility+ profile was removed; `--profile keyless` now exits with an error.+ See README.md and docs/agent-protocol.md.+ Sibyl never calls a generative model or reads subscription credentials. Reasoning+ and any model usage stay in the calling host under its own terms and limits.+ Search keys are optional; Tavily can incur separate charges. Markdown/JSON reports+ need no extra; `[report]` is PDF rendering only, with Pillow >=13.0.0.+ Read `research_plan.questions[].review` for host-authored conclusions and stable+ support/challenge links. Inspect gaps before reuse. Use+ `get_research(passage_ids=[...])` for exact evidence: briefs omit raw text.+ Use `view="sources"` with the current revision to page the full catalog, and+ `view="full"` for the compatible snapshot. Each brief is self-contained.+ `continue_research` and `assess_research` return a `view="update"` for one+ question: new sources with top passages inline (cut ones marked `truncated`),+ retrieval notes, that question's gaps and checks, readiness, next searches and+ `before_answering` steps; after an assessment, `answer_material` lists the+ recorded claims with URLs, counterevidence status and open gaps for the answer.+ Before answering, read at least two independent publishers and one+ counterevidence source; cite URLs, never Sibyl claim or passage IDs.+ 2. Find sources with your own web search when available and read them with+    `continue_research(job_id, url=... or urls=[up to 5], query=..., question_id=...)`;+    read sources are clustered, ranked and assessed like search results. A query+    alone falls back to keyless search, which public search sites often refuse.+    Follow leads, investigate contradictions, and look for original sources.+    passage IDs, atomic claims, a conclusion, and a counterevidence note. Facts cite+    passages and quote exact text from them (unverifiable quotes are rejected);+    reasoning cites basis_claim_ids with a rationale. Recommendations+    declare criteria and their user/host_assumption origin. Unknowns remain+    unresolved with an explanation. See docs/research-claims.md. Run the searches in+    `research_plan.suggested_queries` (each has a kind and the gap-based reason).+    `claim_support` flags facts that rest on one source or only on background

Showing the first 40 of each.

https://yichenwu.dev/llms.txt

− Production AI + full-stack engineer in Greater Vancouver, BC, open to− relocation. Currently a Software Engineer on the Production AI Platform at− DaoAI Robotics.− - DaoAI Robotics, Software Engineer — Production AI Platform (May 2025–Present):−   designed graph-aware Pydantic validation and shipped 20+ reusable workflow−   nodes across detection, tracking, OCR, alerting, and traffic analytics;−   implemented multi-object tracking and production dataset pipelines; delivered−   full-stack generative AI workflows and Celery/Redis orchestration with−   persistent job state, WebSocket progress, retry, and failover; developed a−   Kling V3 structured multi-shot video-generation pipeline.− ## Flagship projects− - Argus: MCP QA that turns coding agents into exploratory test engineers for−   web and native macOS apps. It probes workflows, state, and UX, then re-confirms−   every certified finding from a clean reload and replays it in CI without−   model calls. Product overview: https://yichenwu.dev/projects/argus− - Sibyl: public-beta web research for AI agents via MCP, CLI, and Python.−   Keyless multi-source retrieval ranks passage-level evidence and clusters−   syndicated copies; an optional configured model produces citation-backed−   reports with explicit insufficiency states. Product overview:−   https://yichenwu.dev/projects/sibyl− - Duet: a local Electron collaboration loop for coding tasks. Codex writes in−   an isolated Git worktree, Claude Code reviews from a fresh read-only context,−   and machine checks govern completion. Product overview:−   https://yichenwu.dev/projects/duet−   Source: https://github.com/chriswu727/agent-duet− - Loop: an autonomous software delivery runtime that takes a product brief−   through planning, implementation, testing, repair, and verified handoff. It−   returns reviewable artifacts or Git changes with evidence that the−   user-confirmed acceptance contract passes. Product overview:−   https://yichenwu.dev/projects/loop− - Scalable Starter: a reusable Next.js and FastAPI foundation connecting a web−   app, API, worker, Postgres, and Redis, with contract checks, migrations,−   observability, container builds, and validated Kubernetes overlays.−   https://github.com/chriswu727/scalable-starter− ## Other work− - Clinic Web Platform: an anonymized client delivery for a Vancouver−   medical-aesthetics clinic, combining responsive treatment pages, bilingual−   navigation, custom PHP publishing tools, site search, and automated canonical−   URLs, sitemaps, and redirects.− - Website Knowledge Chat: a private multi-tenant prototype that turns client+ Software Engineer - Production AI Platform at DaoAI Robotics. Home bases in+ Vancouver and Toronto; open to AI or full-stack engineering roles in either city.+ - DaoAI Robotics, Software Engineer - Production AI Platform (May 2025–Present):+   AI Film Platform: built the backend runtime for a Director Agent that turns+   film ideas into editable canvas workflows, with 20 typed tools and server-side+   validation for planning, media generation, and human approval. Decoupled agent+   execution from WebSocket connections using Python asyncio, with Redis Streams+   reconnect replay and PostgreSQL conversation history across API workers.+   Extended Celery/Redis workflows for image, video, and audio generation, adding+   backend status polling and fixing failure-refund gaps and duration-based+   audio billing.+   Video Analytics Platform: implemented the processing logic for 20+ reusable+   nodes and Pydantic workflow validation; built stateful tracking and traffic+   analytics with ByteTrack, OpenCV, NumPy, and SciPy. Removed per-frame Redis serialization from selected+   paths using bounded in-memory state and timestamp guards. In a field deployment,+   the platform supported 48 streams at approximately 5 FPS per stream.+ ## Selected personal projects+ - Argus: a local QA runtime for coding agents, with goal coverage, structured+   symptom checks, and evidence reports. Automatic regression uses built-in+   text and HTTP checks. The 0.7 release validation covers Chromium; live native+   macOS UI and a new paid model benchmark were not exercised.+   Project: https://yichenwu.dev/projects/argus+ - Myth Atlas: a live bilingual mythology website built with AI across code,+   content collection, and selected editorial illustrations. Structured entries,+   bilingual search, side-by-side story comparisons, source references, and+   image provenance support the collection. AI-assisted narratives require source review; generated artwork+   is interpretation, not historical evidence.+   Story: https://yichenwu.dev/projects/myth-atlas+   Live: https://myth-atlas.vercel.app/en+   Source: https://github.com/chriswu727/myth-atlas+ - Sibyl: Deep Research MCP for multi-round investigations, evidence review,+   persistent jobs, and cited reports. The 0.4.0b1 beta focuses on the 12-tool+   keyless host-assisted profile; autonomous research is experimental. Live+   retrieval still misses readiness and stability targets. The beta is published+   on PyPI; install sibyl-research==0.4.0b1 explicitly. Version 0.3.0 remains+   the older stable package.+   Project: https://yichenwu.dev/projects/sibyl+ These are personal projects. The production workflow contributions listed in+ Experience belong to Yichen's employment at DaoAI Robotics.+ ## More engineering work

Showing the first 40 of each.