| name | environmental-preparations |
| description | Running log of the user's environment preferences and configuration changes — CLAUDE.md rules, settings, and conventions. Use to recall how the user likes their environment set up, and append a dated entry whenever you make an environment change (anything that isn't building a skill). |
Purpose
Claude has no memory across sessions. This skill is the durable record of how the user likes the environment configured. Read it when preparing a project's environment; append to it whenever an environment change is made (CLAUDE.md edits, settings, conventions — anything that isn't a skill build).
How to log
Append a new entry under Change log, newest first: a dated heading, what changed, and where it applies.
Change log
2026-06-22 — LLM token/cost discipline as a standing CLAUDE.md rule
- What: Added an "LLM token/cost discipline" rule to global
~/.claude/CLAUDE.md: terse instructions, StructuredOutput over inline schemas, TOON over JSON for payloads, drop empty/no-finding sections, target <15k input tokens — applied when first writing any LLM agent prompt.
- Where:
C:\Users\Amit\.claude\CLAUDE.md (Working conventions section).
- Why: Cost minimization (~$0.2/run, "instructions too long") is a standing constraint for the user, but every new agent/prompt was built verbose and then re-trimmed on request. The rule makes prompts born lean.
2026-06-22 — Standing conventions into global CLAUDE.md (fix skill auto-trigger friction)
- What: Added a "Working conventions (apply unconditionally)" section to the global
~/.claude/CLAUDE.md: (1) the plan.md/tasks.md planning workflow, and (2) always follow python-guidelines before writing Python.
- Where:
C:\Users\Amit\.claude\CLAUDE.md.
- Why: Skills don't always auto-fire, so the user kept re-asking "did you use the skill?" and re-explaining the same conventions each session. CLAUDE.md rules apply unconditionally on every machine that syncs it, so the two most-repeated conventions no longer depend on skill triggering. The
project-planning and python-guidelines skill descriptions are already trigger-tuned; the global rules are the durable backstop.
2026-06-22 — Project planning convention as a project-root CLAUDE.md rule
- What: Added a "Project planning" snippet to drop into each project's root
CLAUDE.md, declaring plan.md as the source of truth and tasks.md as the derived execution breakdown (Milestones, sub-tasks, status table), and instructing session-start to resume at the next Pending/In Progress task.
- Where: The snippet lives in the
project-planning skill ("Project-root CLAUDE.md rule" section), which offers to add it to a project root on first setup.
- Why: The user re-explained this start-of-project ritual nearly every session; encoding it removes the need to re-describe it.
