Skip to content

Add agent skills (humanify, ace-it) + git guidance; repair skill frontmatter#713

Draft
kmontemayor2-sc wants to merge 4 commits into
mainfrom
kmontemayor/humanify-and-git-guidance
Draft

Add agent skills (humanify, ace-it) + git guidance; repair skill frontmatter#713
kmontemayor2-sc wants to merge 4 commits into
mainfrom
kmontemayor/humanify-and-git-guidance

Conversation

@kmontemayor2-sc

@kmontemayor2-sc kmontemayor2-sc commented Jul 21, 2026

Copy link
Copy Markdown
Collaborator

What

Adds and repairs agent guidance under .claude/:

  • .claude/development.md — two additions to the ## Git section: never
    amend an already-pushed commit (add a new commit instead), and propagate fixes
    up a stacked branch chain with git merge (ask first), never by rebasing or
    force-pushing.
  • .claude/skills/humanify/SKILL.md (new) — a skill for writing code
    comments, docstrings, and PR/commit descriptions: comment the why, write the
    standing state rather than the change, turn comma-runs into lists, and trim
    prose without collapsing scannable structure.
  • .claude/skills/ace-it/SKILL.md (new) — a skill for writing fewer,
    real-behavior tests: assert on real outputs instead of mock.call_args, give
    glue code e2e coverage rather than mock-plumbing unit tests, and match test
    count to risk.
  • Frontmatter repairmdformat was corrupting the YAML frontmatter of
    every SKILL.md (folding name/description/argument-hint into a heading
    and escaping brackets). The Makefile now excludes .claude/skills from the
    markdown file set, and the frontmatter of all skills (codex-verify,
    pragmatic, watch-action, humanify, ace-it) is reconstructed as proper
    ----delimited YAML. Real docs under .claude/ (development.md,
    formatting.md) are still formatted.

Why

Codifies recurring guidance so agents write comments and tests that survive
refactors, and stops mdformat from silently breaking skill frontmatter (which
the skill loader needs to parse).

kmontemayor and others added 3 commits July 21, 2026 00:06
Add a `humanify` skill under .claude/skills that guides agents to write
code comments, docstrings, and PR/commit descriptions for a hurried
teammate reading the diff — comment the why at a different level than the
code, write the standing state rather than the change, and trim prose
without collapsing scannable structure.

Extend .claude/development.md with two git rules: never amend an
already-pushed commit (add a new commit instead), and propagate fixes up a
stacked branch chain with `git merge` only after asking, never by rebasing
or force-pushing.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Stop mdformat from corrupting SKILL.md YAML frontmatter by excluding
.claude/skills from the MD_FILES set, then reconstruct proper
----delimited frontmatter (name + description [+ argument-hint]) for every
skill; mdformat had folded these fields into a heading and escaped the
argument-hint brackets.

Refresh the humanify skill to the trimmed revision and add an
"Enumerations become lists" section. Add a new ace-it skill covering how
to write fewer, real-behavior tests and avoid mock-plumbing assertions.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
@kmontemayor2-sc kmontemayor2-sc changed the title Add humanify skill and git history-hygiene guidance for agents Add agent skills (humanify, ace-it) + git guidance; repair skill frontmatter Jul 23, 2026
Replace the inline find excludes with an explicit MD_EXCLUDE_DIRS list so
the set of unformatted directories is easy to read and extend. Adds
experimental sub-projects, plan docs, build/dist output, pytest cache, and
the .sdd/.superpowers agent dirs to the caches/venv/tools/skills already
skipped. Real docs under .claude (development.md, formatting.md) stay
formatted.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
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.

1 participant