tighten-text
Skill2 starsChanged 4 days agoApache-2.0View on GitHub
Tightens English security/assessment prose produced in this repo — threat assessments, the analysis trail, mitigation writeups — into dense expert writing. Cuts filler, hedges, intensifiers, bureaucratese, nominalizations, passive voice, and tangled syntax; never touches substance. Returns edited text, a compression figure, and a rule-linked breakdown. Use on 'tighten', 'make it shorter', 'cut the fluff', 'too much text', 'polish this assessment', 'clean up the writeup'.
In this repository13 other files it gives its agents.
- assess-genai-with-librarySkill
- assessing-genai-securitySkill
- check-styleSkill
- folding-into-the-modelSkill
- humanizerSkill
- keel-designSkill
- assess-genai-with-librarySkill
- assessing-genai-securitySkill
- check-styleSkill
- folding-into-the-modelSkill
- humanizerSkill
- keel-designSkill
- tighten-textSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
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
- Distinctive: line 34
P9. Read-aloud check: Read the whole result. A stumble, an unnatural seam, a lost connective, a dropped nuance — go back.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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.
Files
Tightening security/assessment prose
An execution skill: it edits a given text. Tuned for this repo's output — GenAI threat assessments and library prose — where precision beats brevity. Principles are inline; nothing external is required.
Scope
Does: removes language clutter and de-bullets a chain into connected expert prose, without touching meaning. Typical result — minus 20–40%.
Doesn't, by default: doesn't add facts absent from the source; doesn't change the assessment's structure or verdicts; doesn't invent findings.
The one hard guardrail — never cut substance. Tighten the voice, not the content. The following are load-bearing and MUST survive verbatim in meaning: risk levels ("critical/high/low" are levels, not evaluations to delete), HARD vs SOFT distinctions, GATE ITEMS / stated uncertainty (a hedge that marks an honest missing fact is not filler — keep it), the delta (MITRE ATLAS framing), refutation/"NOT applicable if" reasoning, asset/attacker/surface/scenario specifics, and every domain term. Removing honest uncertainty to sound confident is a defect, not an improvement.
Procedure
Passes in strict order, bottom-up. Because this is security prose, only the precision-safe passes run by default: P1, P3, P5, P7 (and P0/P9/P10). The meaning-touching passes (P2, P4, P6, P8) run only with --hard and always flag rather than guess.
P0. Reader, goal, inventory. Identify the genre and list the protected elements (above). Note the four parameters if they matter: who reads it, what they already know, the goal, the cost of an error. If a needed one is missing, ask once as a block and stop.
P1. Filler & throat-clearing. Stating the obvious, appeals to your own opinion, false candor ("it's worth noting", "to be honest"), needless qualifiers, verbal enumeration, needless "for example", "by the way", empty parentheticals, AI signposting ("Let's dive in", "In summary"). Marker — paired commas/dashes/parentheses at a sentence start.
P3. Clichés & bureaucratese. Compress a cliché to a word; collapse bureaucratic phrasing to a verb; delete time-filler.
P5. Action (nominalizations & passive). Collapse "provide / perform / carry out + a nominalization" into a verb ("performs execution of" → "executes"). Flip passive to active, except when describing a state.
P7. Syntax & de-bullet. Split run-ons with a period; break nested subordination; one idea per sentence (up to three in technical writing). De-bullet: when a bulleted chain encodes step logic (X → Y because Z → impact), render it as one connected scenario sentence, not a field dump — unless the list is a genuine enumeration (then keep it as a list).
P2 / P4 / P6 / P8 (only with --hard). Intensifiers (delete); an evaluation with no supporting fact (drop + [FACT]); fancy word → plain only if meaning fully matches (else [TERM]); vague quantifier → figure ([UNSOURCED] if no source); a string of synonyms → the single precise term (never cut a list that proves the point).
P9. Read-aloud check. Read the whole result. A stumble, an unnatural seam, a lost connective, a dropped nuance — go back. Formally correct is not the same as alive, and shorter is not the same as complete.
P10. Report per the format below.
What we never touch
Numbers, dates, amounts, units; names, titles, orgs, links; domain terms and their exact wording; direct quotes; risk levels and HARD/SOFT tags; GATE ITEMS and stated uncertainty; legally significant wording (may simplify, not drop — [LEGAL]); caveats/warnings that protect the reader (never drop silently — [CAVEAT]); what makes the text actionable (what to do, by when, whom to contact).
Modes
| Flag | Behavior |
| default | P1, P3, P5, P7 + P0/P9/P10 — precision-safe clutter and de-bulleting only |
--hard | plus P2/P4/P6/P8 (meaning-touching), always flagging rather than guessing |
--analyze-only | text not rewritten; returns found problems with quotes; P0 ask skipped |
Report flags
[FACT] evaluation removed, no replacing fact · [UNSOURCED] claim with no source · [TERM] left un-simplified for precision · [LEGAL] legal wording touched · [CAVEAT] formal detail kept · [STRUCTURE] text-level problem, not word-level · [AUDIENCE] call depends on what the reader knows.
Output format
## Result
<edited text, no marks, ready to copy>
## Compression
Was: N words / M chars → Now: N₂ / M₂ (−X% / −Y%)
## Breakdown
| Was | Now | Rule |
|---|---|---|
| <fragment> | <fragment or "deleted"> | <rule> |
## Needs your decision
- [FACT] "<fragment>" — evaluation "<word>" removed, a fact is neededGroup the breakdown by rule with counts ("filler — deleted 11"); show separate lines only for decided edits. If compression came out under 10% — say so and name the reason (already clean / all facts / the problem is structure, not words). Empty after cleaning is a diagnosis, not a failure — say the source had no substance.
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
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.

