Conciseness

The Purpose, Trigger, Process, and Boundaries sections are lean and assume Claude's competence, but roughly half the body ("Compression Rules" and "Pattern", ~55 of 106 lines) restates rules that scripts/compress.py already embeds in its own model prompt ("Preserve ALL URLs exactly", "Preserve ALL headings exactly"), and the code-block/inline-code preservation rule is stated three separate times. It is not 2 because nothing explains general concepts Claude already knows; it is not 4 because the duplication is substantial rather than minor and could be tightened to a short summary.

Actionability

The exact executable command is given ("python3 -m scripts <absolute_filepath>") along with a fallback for locating scripts/__main__.py, concrete before/after compression examples, and explicit extension allow/deny lists. This is copy-paste ready guidance covering the common cases, matching the top anchor rather than 4's "minor gaps".

Workflow Clarity

The Process section gives a clear sequence with explicit validation and a feedback loop ("validate output (no tokens)", "cherry-pick fix with Claude", "retry up to 2 times", "if still failing ... leave original file untouched"), and the destructive overwrite is guarded by an out-of-tree backup. Since validation and error-recovery checkpoints are explicitly present, the destructive-operation cap at 3 does not apply and the content matches the top anchor.

Progressive Disclosure

The body is organized into clear sections and its references to the bundle are real and one level deep (scripts/__main__.py and scripts/compress.py both exist in scripts/), with no nested reference chains. It is not 5 because the ~55-line inline Compression Rules section duplicates script-embedded behavior that could live in a reference file or be summarized, leaving a minor organization gap rather than a fully appropriate split.

Specificity

The description lists several concrete actions — "Compress a memory file such as CLAUDE.md or a todo list into caveman format", "keeping a readable backup", "Sends the file's text to the configured model provider" — with named targets, matching the several-specific-actions anchor. It falls short of 5 because coverage has gaps (no mention of validation, the refusal of sensitive files, or which file types qualify).

Completeness

The "what" is clearly stated (compress memory files to caveman format with a readable backup to save tokens), and the "when" is addressed via "Trigger: /caveman-compress", which counts as explicit trigger guidance so the missing-Use-when cap at 3 does not apply. It stays at 4 rather than 5 because the when-clause names only the slash command and omits natural-language trigger conditions (e.g. "use when the user asks to compress or shorten a memory file").

Trigger Term Quality

Natural terms a user would say are present: "compress", "memory file", "CLAUDE.md", "todo list", "save input tokens", plus the explicit "/caveman-compress" trigger. It is not 5 because common variations like "shorten", "shrink", "reduce context", or other memory-file names (e.g. AGENTS.md, README) are missing.

Distinctiveness Conflict Risk

"Caveman format", "input tokens", and "memory file" carve out a clear niche that is hard to confuse with document or data skills. Minor overlap risk remains because a bare request to "compress a file" could plausibly mean zip/gzip, keeping it below the minimal-conflict anchor 5.