Skip to content
Directory
GitHub Star
ETF AGENT / NOTES

OPEN ETF AGENT / A RUNNABLE FIELD GUIDE

Understand a mechanism.
Run a report you can reopen.

Go beyond architecture essays: run the corresponding implementation, open its report, and trace the code, original sources and execution evidence.

Open source · Synthetic examples · No model key for the default demo

Read the six-layer map by module and capability ​

Select a card to open its corresponding article. Expand, zoom and scroll the map. The map describes bounded Open implementation, not a claim that every capability in the article is public or qualified.

Current scope: v0.30.0 · Fixed navigation source · Code and verification limits

Panorama · select a module to read more

Open the interactive SVG directly

Article links by module (same destinations as the map)

L1 · Inputs & triggers ​

L2 · Identity, session & context ​

L3 · Agent decisions & execution ​

L4 · Tools & domain capabilities ​

L5 · Publication & delivery ​

L6 · User interfaces ​

Dashed cards represent call-sites of existing owners, not additional modules. Article links provide reading navigation only; existing coverage qualification and historical evidence remain unchanged. The GitHub README image is a preview; interactive card navigation is available here.

01 / READ → RUN → INSPECT

Three recent modules to explore.

Full article catalog →

Ordered by module progression, not article publication date. Each essay explains the larger mechanism and marks the smaller public implementation.

New here? Start with the system overview →

Ready to use your own documents? Follow the v0.20 workshop →

02 / TRY IT LOCALLY

Your first report, on your disk.

Requires Node.js 26.8.1–26.x and npm. Installation accesses npm; the installed default example runs offline without a model key.

The CLI prints a new directory under runs/. Open report.md; manifest.json and run.jsonl retain delivery and execution records. This is a synthetic example, not live fund research.

Setup details & more examples ↗
Terminal
git clone https://github.com/hpoenixf/open-etf-agent.git
cd open-etf-agent
npm ci --ignore-scripts --no-audit --no-fund
npm run build
npm run demo -- --case document-compare

03 / THE BUILD LOG

One increment. A clearer system.

Source lineage ↗

Fixed development snapshots, not a release registry. Open any stage to inspect its original code; each new stage has one distinct delivery focus.

v0.14 read-only Memory: fixed source and runnable example ↗

  1. 0.30Control the original Child lifecycleSpawn · inspect · cancel · immutable observations · cold recovery
  2. 0.29Continue an explicit conversationMessage identity · fresh grants · bounded historical context
  3. 0.28Review a frozen reportAdvisory findings · original sources · cold readback
View all versions · 30

Complete development history

  1. 0.1First readable reportRuntime · Gateway · Tool Host
  2. 0.2Methods meet toolsSkill · MCP · SSE
  3. 0.3Cooperate within boundsChild · Hook · shared budget
  4. 0.4Keep the originalResult · Context · Retrieval
  5. 0.5Notes are not factsCompaction · explicit Memory
  6. 0.6Deliver by identityArtifact · Revision · Reader
  7. 0.7Contribute and recoverPlugin · SQLite · lease / fence
  8. 0.8Commit the answer bodyPublication · revision CAS
  9. 0.9Continue the same RunSubmit · Worker · readback
  10. 0.10One result, actual consumersResult / Reference · report · CSV
  11. 0.11Read the original PDFLocal text layer · physical page · original bytes
  12. 0.12Keep task time fixedTemporal · IANA zone · original task date
  13. 0.13Adopt the original replyOrdinary response · receipt · same-Run adoption
  14. 0.14Read existing MemoryMemory · current revision · actual request
  15. 0.15Check sources before handoffSpecialist Child · exact evidence · Main
  16. 0.16Recheck eligibility on readRevocable source · Publication · Reader
  17. 0.17Share admission capacityShared Admission · original request
  18. 0.18Wake one confirmed occurrenceSchedule · explicit wake · same Run
  19. 0.19Compose existing contractsOriginal owners · explicit inputs · bounded summary
  20. 0.20Run your own research projectLocal sources · same-Run recovery · layered evidence
  21. 0.21Return answers to the same taskSequential questions · durable receipt · explicit adoption
  22. 0.22Delegate bounded read-only workSerial Child · scoped sources · original Main
  23. 0.23Retrieve admitted evidenceKnowledge · lexical retrieval · original source
  24. 0.24Keep project capacity across revisionsEndpoint identity · original tickets · project-local scope
  25. 0.25Confirm and run one scheduled taskOnce schedule · explicit due work · original lifecycle
  26. 0.26Acquire bounded Web evidencePublic static pages · originals · source-backed readback
  27. 0.27Freeze project specialist methodsExact revisions · ordered presets · original Child
  28. 0.28Review a frozen reportAdvisory findings · original sources · cold readback
  29. 0.29Continue an explicit conversationMessage identity · fresh grants · bounded historical context
  30. 0.30Control the original Child lifecycleSpawn · inspect · cancel · immutable observations · cold recovery

GitHub Releases ↗

04 / WHAT TO LOOK FOR

Follow the result, not just the diagram.

Read the original

Keep source identity distinct from summaries and working notes.

Continue the same Run

Inspect explicit checkpoints and bounded recovery rather than assume a retry is safe.

Reopen the delivery

Trace report files and artifact revisions back to their recorded execution.

05 / THE FULL READING MAP

Read the idea. Inspect the boundary.

Full catalog →

Start here: the v0.10 system overview →
Then revisit the v0.1 origins: the first readable report →

Module essays follow the existing catalog; related topics retain their own categories. A receipt proves only the named slice. Broader essays and later additions remain distinct from implemented coverage.

Expand modules and related reading · 50

01L1 · Task Input and Triggers

Limited target · receipt available

Scheduled Research Beyond a Conversation

Explicit submit and separate work process read back the original Run; no schedule trigger, Task orchestration, notification or foreground concurrency.

One confirmed due-time Schedule API occurrence with explicit wake; ten successful tasks retain the same Run through Worker, Admission, Publication and read-only Session inspection. Separate fixture CLI; no background or recurring service.

Inspect additional target ↗ · since v0.18
Inspect target ↗ · since v0.9中文原文

02L2 · Identity, Sessions and Context

03L3 · Agent Decisions and Execution

04L4 · Tools and Domain Capabilities

05L5 · Result Publication and Delivery

06L6 · User Interface

07Related reading · Architecture and Systems Thinking / Deep and Continuous Research

08Related reading · Architecture and Systems Thinking / Subagents and Experts

09Related reading · Architecture and Systems Thinking / System Boundaries and Evolution

010Related reading · AI Agent Engineering Practice / Real Validation and Task Convergence

Engineering notes, evidence and boundaries. · Contact: hpoenixf@foxmail.com · WeChat: hpoenixf0510