From compress
Compress agent/skill definitions using math/logic notation. Triggers: "compress" | "compress skill" | "compress agent" | "compress context" | "shorten this" | "make it formal" | "use formal notation" | "make this more compact" | "compress this".
How this skill is triggered — by the user, by Claude, or both
Slash command
/compress:compressThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Formal notation rewrite — reduce tokens, preserve semantics.
Formal notation rewrite — reduce tokens, preserve semantics.
/compress code-review → .claude/skills/code-review/SKILL.md
/compress fixer → .claude/agents/fixer.md
/compress file.md → direct path
∀ all | ∃/∄ exists | ∈/∉ member | ∧/∨ and/or | ¬ not | → then | ⟺ iff | ∅ empty | ∩/∪ intersect/union | ⊂ subset | ∥ parallel | |X| count | :=/← assign | { } scoped block | ; sequence | () params | ↦ maps to
1 — Resolve: Parse $ARGUMENTS: *.md → direct | agent name → .claude/agents/<name>.md | skill name → .claude/skills/<name>/SKILL.md | ∅ → → DP(B) file ∃ → read. ∄ → halt.
2 — Analyze: Read target. Identify: repeated nouns (≥3×) | verbose conditionals | iteration prose | magic numbers | redundant examples | filler. L_before := line count (excl. frontmatter).
3 — Transform (R1–R10):
Let: block (after title). Lowercase, mnemonic. Template:
Let:
φ := set of all findings
γ(f) ∈ [0,100] ∩ ℤ — confidence
τ := 80 — threshold
pred(x) ⟺ A ∧ B ∧ C∀ x ∈ Y: | "if any" → ∃ x: | "exists" → X ∃ →X → YO_name { step₁; step₂; … } → outputF(x, y)¬compress: frontmatter | code blocks | $ARGUMENTS | file paths | tool names | safety rules | table structure
4 — Present: Show L_before → L_after (N%) + substitutions. → DP(A) Yes | Preview | Adjust. Preview → show, re-ask. Adjust → apply, re-present.
5 — Write: Write file. Verify: frontmatter ∧ $ARGUMENTS ∧ safety rules ∧ ¬semantic loss. Report L + %.
| Scenario | Behavior |
|---|---|
| Already formal | "already compressed", tweaks only |
| L < 30 | Warn, proceed if confirmed |
| Agent (¬skill) | Preserve agent frontmatter |
| No repeated concepts | Skip R1, apply R2–R10 |
| User rejects | Halt |
| Mixed prose + code | Prose only |
$ARGUMENTS for skills$ARGUMENTS
npx claudepluginhub roxabi/roxabi-plugins --plugin compressCompresses prompts and skills into minimal goal-focused instructions by dropping redundant details, trusting model capabilities, and maximizing action space. Use to condense or minimize prompts.
Optimizes AI-facing instructions for subagents, CLAUDE.md, rules, skills, agents by removing discoverable data, explained knowledge, invented constraints, stale facts. Use for authoring or auditing bloat.
Compresses SPEC.md and spec-adjacent writes into a concise caveman encoding, cutting tokens ~75% while preserving precision. Activates on SPEC.md writes or user requests for brevity.