Skip to content

docs: condense CLAUDE.md to skill-first kernel#47

Open
gaberger wants to merge 3 commits intomainfrom
claude/condense-claude-md-rgEra
Open

docs: condense CLAUDE.md to skill-first kernel#47
gaberger wants to merge 3 commits intomainfrom
claude/condense-claude-md-rgEra

Conversation

@gaberger
Copy link
Copy Markdown
Owner

CLAUDE.md shrinks from 623 → 93 lines. It now holds only the
non-negotiable rules, tool precedence, architecture invariants,
build one-liners, and a skill-map for discovery. Everything else
moves into docs/reference/ (9 files, ~560 lines) backing the
/hex-* skills as progressive-disclosure material agents pull in
on demand rather than loading every session.

Also allowlist docs/reference/ in .gitignore.

claude added 3 commits April 17, 2026 08:54
CLAUDE.md shrinks from 623 → 93 lines. It now holds only the
non-negotiable rules, tool precedence, architecture invariants,
build one-liners, and a skill-map for discovery. Everything else
moves into docs/reference/ (9 files, ~560 lines) backing the
/hex-* skills as progressive-disclosure material agents pull in
on demand rather than loading every session.

Also allowlist docs/reference/ in .gitignore.
- ADR-2604170900 (Accepted): CLAUDE.md becomes skill-first kernel;
  623→~90 lines; reference material moves to docs/reference/ backing
  /hex-* skills. Shipped in commit 0ccc02a.
- ADR-2604170905 (Proposed): /hex-orient skill for fresh-session
  grounding without reinflating CLAUDE.md.
- ADR-2604170910 (Proposed): absorb task-tier classification into
  /hex-workplan rather than create a separate /hex-classify skill.
5 phases / 10 tasks: author skill → register in skill-map +
catalog → rebuild hex-cli+hex-nexus → verify deployment via
hex init → commit + flip ADR to Accepted. Not executed —
draft only.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants