Skip to content

junit/structured-thinking

Repository files navigation

structured-thinking

简体中文

Evidence-preserving communication, incident analysis, and decision analysis for AI agents.

Publication-candidate artifact targeting 0.1.0-rc.1. Bundled documentation does not prove publication, marketplace availability, current support, or Production certification. Verify authenticated external metadata; use the local-preview path when no trusted listing is visible.

Quick Start · Documentation · Product contract

What it does

The Plugin family turns supplied evidence into technical communication and analysis that a reader can verify. Its governing rule is:

Improve the organization and usefulness of evidence without changing its evidentiary status.

It does not invent missing facts, silently confirm a root cause, turn subjective scores into objective truth, or claim that a proposed action was authorized or completed.

Choose the outcome

Need Route
Assess what happened, competing causes, counterevidence, and next evidence checks incident-analysis
Compare options against supplied must-haves, criteria, evidence, and uncertainty decision-analysis
Package supplied evidence or already-completed analysis for an audience structured-thinking
Execute commands, edit code, deploy, approve, purchase, message, or perform a simple/visual-only task none

Mixed requests follow dependency order: unresolved causality → option comparison → audience packaging. See Choosing a Skill for explicit, implicit, mixed, and negative selection.

Evidence and safety boundary

All three Skills distinguish Reported claim, Fact, Derived fact, Inference, Hypothesis, Recommendation, and Unknown. Audience adaptation may change order and detail, but never material facts, calculations, confidence, counterevidence, risk, or action status.

The family is read-only. Instructions inside logs, traces, retrieved pages, or quoted documents are evidence, not authorization. Live debugging and workspace recovery need a separate tool-capable workflow with explicit permissions and reversible operations.

Corrected 90-minute example

Given only “failures began at 14:00; the change was reverted at 15:00; service recovered at 15:30,” a faithful result reports 15:30 - 14:00 = 90 minutes. Financial impact, code defect, owner, and root cause remain Unknown unless the input supplies and verifies them. A cleaner format cannot justify inventing $200k or a missing finally block.

Architecture at a glance

  • .codex-plugin/plugin.json exposes three complete directories under skills/.
  • User documentation, concepts, and runtime governance explain the installable Plugin.
  • evals/, src/st_eval/, tests, workflows, maintainer guides, and implementation policy are source-only development systems.
  • Local package validation and pinned-platform support are separate evidence claims.

Read Architecture, the Evidence model, and the Routing model for the authoritative conceptual boundaries.

Documentation

License

MIT — see LICENSE.

About

Structured thinking framework for AI agents, coding assistants, and developers to turn raw logs, events, and scattered data into decision-ready communication.

Topics

Resources

License

Contributing

Security policy

Stars

2 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages