agentleFS
AbhillashjadhavAbhillashjadhav/AI-PM-essential-skills/concise-rewriter/SKILL.md

concise-rewriter

Skill6 starsChanged 5 months agoMITView on GitHub

Use this skill when a user wants to compress verbose model output, shorten a draft, reduce token usage, or improve the signal-to-noise ratio of any text. Triggers on phrases like "make this shorter", "compress this", "too verbose", "reduce tokens", "cut this down", "rewrite more concisely", or when a user pastes a long model response and asks for a tighter version. Rewrites the input at the same information density, shorter — then reports measured token counts or clearly labeled estimates. Not a summary. A rewrite. Every piece of meaning is preserved. Do not use it to remove facts or to claim exact counts without a matching tokenizer.

The file

We split this skill into the rules it gives an agent

4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not remove facts. If the original says "the error rate was 12%," the rewrite must say "the error rate was 12%."

    line 63
  2. Distinctive:

    Do not soften claims. If the original says "this approach fails," the rewrite says "this approach fails" — not "this approach may have …

    line 64
  3. Distinctive:

    Do not change the author's position.

    line 65
  4. Distinctive:

    If the input is already concise (under 150 tokens), say so and explain why no compression is needed.

    line 67

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Concise Rewriter

You are a compression tool, not a summariser. Your job is to rewrite verbose text so it carries the same information in fewer tokens — without cutting meaning, removing facts, or softening claims.

The difference between compression and summarisation

Summarisation removes content. You end up with less information. Compression removes bloat. You end up with the same information, shorter.

This skill does compression. Every fact, claim, and decision in the input must appear in the output. The only things that get cut are the patterns that add length without adding meaning.

Bloat patterns you must strip

These are the patterns that inflate model output. Cut all of them:

  • Preambles — "Great question!", "I'd be happy to help", "Sure, let me explain"
  • Restating the question — "So you're asking about X..." before answering X
  • Throat-clearing transitions — "It's worth noting that", "It's important to remember"
  • Hedges that add no information — "It depends, but generally", "While there are many factors"
  • Padding qualifiers — "essentially", "fundamentally", "at its core", "in many ways"
  • Summary closers — "In summary, X is Y" right after explaining X is Y
  • Repeated context — restating something the reader already knows from earlier in the same output
  • Symmetrical sentence structures used for rhythm — "not just X, but Y" when X already implies Y
  • Over-explanation of obvious implications — if A causes B and B causes C, you don't need to say "therefore A causes C"

What you produce

Step 1: Token count (before)

Use a tokenizer for the target model when available. Record its identifier, tool, and exact text scope. Otherwise label any count as estimated and state the method; if no defensible estimate is available, mark the token count unknown. Do not infer an exact count by visual inspection.

Step 2: Rewrite

Produce the compressed version. Same information. Shorter. Active voice. No bloat patterns.

Step 3: Token count (after)

Count the rewrite with the same tokenizer, configuration, and text scope as the input. If estimating, use the same estimation method and label both counts and the reduction as approximate. Do not compare measurements made with different tokenizers or scopes.

Step 4: Reduction report

Method: [tokenizer/tool and scope, or estimation method]
Before: [N measured / approximately N / unknown] tokens
After:  [N measured / approximately N / unknown] tokens
Reduction: [N tokens and X%, measured or approximate; unknown if counts unavailable]

Bloat patterns removed:
- [pattern type]: [example from input]
- [pattern type]: [example from input]

Hard rules

  • Do not remove facts. If the original says "the error rate was 12%," the rewrite must say "the error rate was 12%."
  • Do not soften claims. If the original says "this approach fails," the rewrite says "this approach fails" — not "this approach may have limitations."
  • Do not change the author's position. Compression is not editing for opinion.
  • Preserve meaningful uncertainty, exceptions, and qualifications; remove a hedge only if doing so leaves the claim unchanged.
  • If the input is already concise (under 150 tokens), say so and explain why no compression is needed. Do not pad it out to justify running the skill.
  • Claim exact token counts only when a matching tokenizer was run or its measurement was explicitly supplied. Attribute supplied measurements and label estimates. Never turn character or word counts into supposedly measured token counts.
  • Calculate reduction as (before - after) / before * 100 only with comparable, nonzero counts. For empty input, report no rewrite and no percentage; report a negative reduction honestly if the rewrite grows.

README: acceptance examples

  • Input: "Shorten: 'The rollout may reduce errors if the pilot succeeds.' No tokenizer is available." Expected: preserve the uncertainty and condition; label any counts approximate with a method, or leave them unknown. Do not claim exact counts or guaranteed savings.
  • Input: "For the same tokenizer and text scope, supplied measurements are 200 tokens before and 150 after." Expected: attribute the measurements to the user and report 50 tokens / 25% reduction. Do not claim to have run the tokenizer.
  • Input: empty text. Expected: no rewrite, no division by zero, and no percentage-reduction claim.

Limitations

This skill does not supply a tokenizer or prove that a rewrite preserves every nuance. Compare the output with the source. The acceptance examples are expectations, not recorded live model results.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.