context-probe-plugin

AI-Assisted Design Measurement Platform

Japanese version: README.ja.md

This is a docs-first repository for measuring design quality with evidence, using AI-assisted extraction together with deterministic analyzers. The current implementation covers domain-design and architecture-design evaluation, and the documentation is split so future packs can extend the same measurement model without changing the reading model.

Japanese documents are the primary source of truth. English documents mirror the same structure and document roles.

Start With Skills

This repository treats skill-based context-probe usage as the main path. Direct CLI execution is still supported, but it is the secondary entry point below.

• Claude Code plugin details: plugin/README.md
• Available workflows:
  • context-probe-inputs-workflow: prepare and curate YAML inputs for a repository
  • context-probe-analysis-workflow: run measurement, reporting, gate checks, and unknown review after inputs exist

In Claude Code, this repository already ships plugin settings. After trusting the repository, collaborators should be prompted to install and enable the marketplace and plugin automatically. If needed, the manual setup is:

/plugin marketplace add CAPHTECH/context-probe
/plugin install context-probe-plugin@context-probe-marketplace
/plugin enable context-probe-plugin@context-probe-marketplace

Then ask for flows such as:

• Apply context-probe to this repository
• Run context-probe on this repo
• Analyze this context-probe result
• Inspect the unknowns in this score

Use the inputs workflow first when the repository still needs maintained YAML. Use the analysis workflow after curated inputs already exist and you want an actual measurement or investigation loop.

For Codex, use the skills directly from plugin/skills/context-probe-inputs-workflow/SKILL.md and plugin/skills/context-probe-analysis-workflow/SKILL.md. Those workflows assume the published CLI entry point, so command recipes use npx context-probe ... rather than npm run dev -- ... from this repository checkout.

Start Reading

1. docs/README.md
2. docs/guides/user-guide.md
3. docs/guides/repo-apply-playbook.md
4. docs/concepts/measurement-model.md
5. docs/reference/domain-design-metrics.md
6. docs/reference/architecture-design-metrics.md
7. docs/implementation/runtime-and-commands.md

Documentation Map

• docs/README.md: documentation index
• docs/guides/user-guide.md: quickest path for first-time CLI use
• docs/guides/repo-apply-playbook.md: practical path for applying context-probe to an existing repo
• docs/concepts/: conceptual specifications and measurement model
• docs/reference/: how to interpret metrics and summary scores
• docs/implementation/: how the current CLI computes and reports the metrics
• docs/operations/: policy, CI, and collector operation guidance
• docs/roadmap/: phased rollout and experimental notes

First Commands To Learn

• score.compute
• report.generate
• gate.evaluate
• review.list_unknowns

Core Principles

• Use AI as an evidence extractor and ambiguity reducer, not as the scorer.
• Compute scores through fixed formulas and deterministic analysis.
• Attach evidence, confidence, unknowns, and provenance to every metric.
• Prefer candidate comparison and time-series comparison over cross-organization ranking.
• Add new evaluation domains as packs on top of a shared measurement foundation.

Current Implementation

• TypeScript / Node CLI implementation is available.
• Phase 1 capabilities include dependency analysis, boundary-leak detection, evolutionary locality, score computation, reporting, and gate evaluation.
• Phase 2 entry points include external CLI extractors for doc.extract_*, evidence-backed term links via trace.*, and review log support via review.resolve.
• Pack boundaries for domain_design and architecture_design are already present for future expansion.

Quick Start

Choose the shortest path for your goal:

• First-time CLI use: docs/guides/user-guide.md
• Apply context-probe to an existing repo: docs/guides/repo-apply-playbook.md
• Run and maintain this repository's self-measurement: docs/operations/self-measurement-runbook.md