.gitignore

# Built apps are artifacts; never tracked anywhere in this repo.
builds/

# Course folders are tracked, but only the canonical spec, an optional
# variant spec, and the source materials ship: courses/<name>/spec.md,
# courses/<name>/spec.coding-agent-lab.md, and courses/<name>/materials/.
# The variant file is a tracked exploration copy of the canonical spec
# (e.g. testing a Decision Ledger change) — same name across every course,
# so this one negation covers all of them without a per-experiment edit.
# Everything else per course (builds/, evals/, scratch) stays local.
# Ignoring at depth two (courses/*/*) instead of courses/* lets the
# negations work — git can't re-include files whose parent directory is ignored.
courses/*/*
!courses/*/spec.md
!courses/*/materials/
!courses/*/spec.coding-agent-lab.md

# Personal scratch file — kept on disk, not version-controlled.
notes.md

# skill-creator eval artifacts, both the run workspace (per-run specs,
# gradings, benchmark, viewer HTML, scratch course copies that include course
# materials — see courses/* above) and the per-skill evals/ test cases.
# These are point-in-time records of a manual eval, not part of the harness:
# the durable spec-quality contract lives in the generation guide (its §3/§13/
# §14), so a tracked eval file would just be a second source of truth that
# drifts. skill-creator regenerates evals/ into the workspace when re-run.
.claude/skills/*-workspace/
.claude/skills/*/evals/

# Per-user Claude Code permissions allowlist; shared settings live in
# .claude/settings.json (the project-shared one).
.claude/settings.local.json

# macOS noise
.DS_Store

# IDE noise
.idea/

# Python bytecode (from running skill scripts)
__pycache__/
*.pyc