Productivity

Trim

Reduces an always-on LifeOS context file that has grown too big via a human-gated pass — deterministic GC of stale entries first, then semantic merges and relocations — never dropping a directive and committing every change reversibly

01
Workflow
00
Tools
00
References
08
Triggers

Key Points

Not for general code refactoring, trimming video or audio media (use AudioEditor for audio files), or removing AI writing patterns from prose

Inside the Skill

The thinking, frameworks, and architecture that distinguish this skill from a generic version of the same task.

Quick Reference

  • Target resolution: a bare name (OPERATIONAL_RULES) resolves against the always-on set — the system prompt, CLAUDE.md, its @-imports, and the hook-injected memory files. No arg → wc -c that set and take the largest.
  • Order is safest-first: (1) show state, (2) deterministic GC (zero-risk), (3) semantic trims (human-gated), (4) safety gate, (5) re-measure. Full steps: Workflows/Trim.md.
  • One tool it orchestrates — never reimplement: LIFEOS/TOOLS/ProposalGC.ts (removes superseded/duplicate/absorbed entries). Sizes come from wc -c.
  • Three semantic moves: MERGE overlapping rules, TIGHTEN verbose ones, RELOCATE rarely-used detail to an on-demand reference (leave a stub + pointer).
  • The invariant: a trim never drops a distinct directive. If a merge would, keep the original.

Gotchas

  • USER files commit to the USER_DATA repo, not ~/.claude. LIFEOS/USER/** (OPERATIONAL_RULES, PROJECTS, the identity files) is a symlink into a separate private repo. Commit with git -C ~/.config/LIFEOS/USER …. A ~/.claude commit captures nothing under LIFEOS/USER/ — a false safety net.
  • The file can change mid-edit. The autonomic memory loop appends proposals to these files while you work. If a Write/Edit reports "modified since read", RE-READ before writing — a concurrent correction may have landed (this is how a real deploy-command fix was nearly reverted). Never write from a stale read.
  • Semantic merges must never drop a directive. Before applying any merge/tighten, confirm every proper noun, path, tool name, and imperative from the originals survives in the result. If one is missing, the merge is wrong — keep the original. Deterministic GC (superseded/dup/absorbed) is always safe; semantic edits are the risky class.
  • bun/bunx only, never npm/npx.
  • Deterministic first, always. Run ProposalGC before proposing any semantic edit — the free, zero-risk removals often clear enough that no judgment-call edit is needed.

Examples

/trim OPERATIONAL_RULES
# → shows 40,456 B → ProposalGC dry-run (0 removable) → ranks semantic trims (fold the
# 40-entry proposal tail, relocate skill-scoped directives to a reference)
# → applies approved ones behind the safety gate → commits to USER_DATA → 28,043 B (−31%)

/trim
# → no arg: wc -c the always-on set, take the largest, then the same walkthrough

Workflows & Routing · 1

Each workflow is one job the skill runs. The trigger phrases route your request to the right one — this is the skill's routing table.

  1. 01
    Trim Workflows/Trim.md

How to Invoke

Say any of these to your DA and LifeOS activates the Trim skill automatically:

  • "/trim"
  • "trim the context"
  • "trim OPERATIONAL_RULES"
  • "this file is too big"
  • "reduce a doctrine file"
  • "prune an always-loaded file"
  • "fold the proposal inbox"
  • "shrink CLAUDE.md or DA_IDENTITY"

Or invoke explicitly:

Skill("Trim")

Want LifeOS to do this for you?

Install LifeOS on your machine — your DA gets the Trim skill plus 55 others, all hooked into one Life OS.