From claude-rudder
Reference and worked example for the CONTEXT.md / CLAUDE.md two-file workflow. Use when the user asks "what is CONTEXT.md", how the human-context vs agent-briefing split works, or wants a copyable example pair to seed a new repo. Pairs with the operational skills (context-to-claude, claude-to-context, add-to-context, remove-from-context, manual-context-update, chunk-repo-context) — this skill explains the model; those skills do the work.
npx claudepluginhub danielrosehill/claude-code-plugins --plugin claude-rudderThis skill uses the workspace's default tool permissions.
Two-file pattern for managing project context in Claude Code. This skill is reference material — it documents the model and ships a worked example. For operations on these files, use the dedicated skills (listed at the end).
Mandates invoking relevant skills via tools before any response in coding sessions. Covers access, priorities, and adaptations for Claude Code, Copilot CLI, Gemini CLI.
Share bugs, ideas, or general feedback.
Two-file pattern for managing project context in Claude Code. This skill is reference material — it documents the model and ships a worked example. For operations on these files, use the dedicated skills (listed at the end).
Humans express intent best in conversational, narrative form. Agents work best on structured, concise briefings. LLMs are very good at converting between them. The pattern leans into that asymmetry.
CONTEXT.md — human-authored scratchpad
CLAUDE.mdCLAUDE.md — agent-ready briefing
CONTEXT.md for deeper backgroundThe two files are kept in sync by deliberate sync points, not by trying to write a single document that serves both audiences.
Good fit:
Skip this pattern when:
CLAUDE.md is fine aloneWhen CONTEXT.md grows past ~500–1000 lines, split it into context-data/ with topic files and a README index. CLAUDE.md then references context-data/ instead of (or alongside) CONTEXT.md. The chunk-repo-context skill handles the split.
examples/web-app/ in this skill ships a paired CONTEXT.md + CLAUDE.md for a hypothetical reading-tracker web app, plus a README.md annotating what to notice in each file.
To seed a new repo from it:
SKILL_DIR=$(dirname "$0") # or wherever this skill is mounted
cp "$SKILL_DIR/examples/web-app/CONTEXT.md" ./CONTEXT.md
cp "$SKILL_DIR/examples/web-app/CLAUDE.md" ./CLAUDE.md
# then edit both to match the actual project
Read both files side-by-side first — the point is to internalize the difference in voice and structure between the two, not to keep the reading-tracker content.
claude-rudder ships two hooks (in hooks/hooks.json) that reinforce the pattern automatically:
CONTEXT.md or context-data/ and reminds the agent they existNo setup required — they activate when the plugin is installed.
| Need | Skill |
|---|---|
Generate CLAUDE.md from CONTEXT.md | context-to-claude |
Pull info from CLAUDE.md back into CONTEXT.md | claude-to-context |
| Add a new piece of context cleanly | add-to-context |
| Remove outdated/wrong context | remove-from-context |
| Guided coordinated manual edit of both files | manual-context-update |
Split a large CONTEXT.md into context-data/ | chunk-repo-context |
| Run a structured audit on existing context files | context-health-check |
Prune a bloated CLAUDE.md to essentials | chunk-claude / chunk-claude-md |
Writing CONTEXT.md:
Maintaining CLAUDE.md:
CONTEXT.mdCONTEXT.md so the agent knows deeper context existsCONTEXT.md shifts substantiallySync cadence:
CONTEXT.md → run /context-to-claude/claude-to-context to capture them in human form/remove-from-context rather than letting it rotThis skill folds in the conceptual content of the standalone Claude-Code-Context-Toolkit repo. The operational commands from that repo were already present in claude-rudder; what's preserved here is the workflow articulation and the worked example.