@recallnet/codecontext-claude-plugin

Claude Code plugin that surfaces inline @context annotations automatically during reads and edits. The goal is to keep critical constraints, risks, and historical decisions in the agent loop without relying on manual reminder commands.

Install

Install the npm package:

npm install @recallnet/codecontext-claude-plugin

Then point Claude Code at the installed plugin directory:

claude --plugin-dir ./node_modules/@recallnet/codecontext-claude-plugin

What it does

The plugin adds three native hooks:

| Hook | Event | Behavior | | --------------- | --------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------- | | on-read | PostToolUse -> Read | When Claude reads a file with @context tags, a priority-tiered reminder is injected into the conversation so critical annotations surface first. | | on-edit-guard | PreToolUse -> Edit | Write | Before Claude edits a file with !critical annotations, an advisory warning reminds it about those constraints. | | on-stop-check | Stop | When Claude finishes responding, staged files are checked for stale context before commit. |

It also installs three slash commands:

| Command | Purpose | | ----------------------- | ---------------------------------------------------------------- | | /context-scope <file> | Pre-edit briefing: show all annotations on a file by priority | | /context-diff <file> | Post-edit check: show which annotations are affected by a change | | /context-staged | Staged-file staleness check |

What the reminders look like

Priority-tiered steering groups annotations by urgency:

-- MUST-READ (critical) --
L42  @context decision:constraint !critical -- Strict > avoids duplicate processing

-- WARNING (high) --
L78  @context risk:security !high -- Rate limit bypass if cache evicted

-- INFO --
L103 @context history -- Migrated from v1 API

When an annotation links to a Claude skill via {@link file:.claude/skills/.../SKILL.md}, the hook explicitly tells Claude to load that skill before editing.

Why use this instead of manual CLI calls

Without the plugin, the normal loop is manual:

1. run a scope command before editing
2. remember the critical constraints while editing
3. run a diff or staged check afterward

With the plugin:

1. Claude reads a file and gets the @context reminder automatically
2. Claude starts an edit and gets warned about !critical constraints
3. Claude stops and the plugin checks staged files for stale context

The plugin uses the same codecontext CLI and parser under the hood, but keeps the workflow inside Claude Code instead of depending on manual operator steps.

Related packages

• @recallnet/codecontext-cli: CLI for scope, diff, staged, and report flows
• @recallnet/codecontext-parser: parser and staleness logic
• @recallnet/codecontext-formatter: reminder formatting and cooldown helpers

Docs

• Repo: https://github.com/recallnet/codecontext
• Full README: https://github.com/recallnet/codecontext/blob/main/README.md