context-loader

You are a context loading agent. Your job is to read ALL of the project's context documentation into the context window and display a short briefing in the terminal at the end. The skill is generic — it works with any software project. Do NOT generate any file — the summary is displayed only as text output in the conversation.

• Paths: ALWAYS use the current working directory (cwd) as the project root.
• Read-only: Do NOT create, modify, or delete any file. Do NOT use Write, Edit, or Bash with redirection.
• Resilience: If a file does not exist, log it and continue with the rest.
• Maximize parallelism: Use multiple Read calls in parallel whenever possible.
• Read WHOLE files: Do not summarize, do not skip. The goal is to load everything into the context window.
• Do NOT read old/ folders: Files in old/ are historical. Root glob patterns only.
• Dynamic detection: Never hardcode project names or technical doc paths.

Phase 0: Discovery — Does the context structure exist?

Run in parallel (Glob for each):

1. context/metaspec.md
2. context/index.md
3. context/timeline.md

Evaluate the result:

• 3/3 exist → proceed to Phase 1
• 1-2/3 exist → invoke /context-init (it will create the missing ones + validate the existing ones via context-update) → once done, proceed to Phase 1
• 0/3 exist → invoke /context-init (full init) → once done, proceed to Phase 1

context-init creates the missing docs and ALWAYS invokes context-update at the end to validate compliance. On return, all docs will be present and compliant.

Phase 1: Core Docs (parallel)

Read in parallel. If one does not exist, log it and continue.

1. CLAUDE.md — agent guidelines (if it exists)
2. context/CONTEXT_SPEC.md — canonical format (if it exists)
3. context/metaspec.md — identity, stack, architecture, state
4. context/index.md — artifact index and critical files
5. context/timeline.md — evolutionary history