OpenSkillIndex← back to search
claude-code

Talk-normal talk-normal

Stop LLM slop. A curated system prompt that cuts verbose, corporate-sounding LLM output by 56-73% (measured) while preserving information. Works bilingually (English + Chinese). Installs into your AGENTS.md as an always-on behavior modifier.

install
source · Clone the upstream repo
git clone https://github.com/hexiecs/talk-normal
Claude Code · Install into ~/.claude/skills/
T=$(mktemp -d) && git clone --depth=1 https://github.com/hexiecs/talk-normal "$T" && mkdir -p ~/.claude/skills && cp -r "$T/skill-hermes" ~/.claude/skills/hexiecs-talk-normal-talk-normal && rm -rf "$T"
manifest: skill-hermes/SKILL.md
source content

talk-normal

A curated system prompt that stops your LLM from writing like a LinkedIn post. Measured: 73% character reduction on GPT-4o-mini, 72% on GPT-5.4, across 10 prompts in English and Chinese, without losing information.

When to Use

Run the installer once per workspace to make your Hermes agent permanently less verbose. This is not a per-turn skill — it is a one-time installer that injects always-on rules into your project's context file.

Procedure

To install or update talk-normal in the current workspace:

bash install.sh

The script auto-detects your workspace config file (

.hermes.md
, 
HERMES.md
, or 
AGENTS.md
) and injects the prompt between 
# --- talk-normal BEGIN ---
and 
# --- talk-normal END ---
markers. The installer is idempotent: running it again replaces the existing talk-normal block in place with the latest rules. Nothing else in your config file is touched.

To remove:

bash install.sh --uninstall

Start a new Hermes session to take effect.

What gets installed

The contents of 

prompt.md
get injected into your workspace config file. The rules target the specific slop patterns that make LLM output sound corporate and padded:

  • Filler and hedging — banned opening phrases, no throat-clearing, no restating the question
  • Structural discipline — lead with the answer, match depth to complexity, bullets only when genuinely parallel
  • Closing patterns — no summary stamps, no hypothetical follow-up offers, no conditional next-step menus
  • Rhetorical tics — no negation-based contrastive framing, no plain-language restatements
  • Shape of comparisons — give a recommendation with brief reasoning, cap pros/cons lists

Verification

After installing, start a new Hermes session and ask a simple question like "What is Python?". The response should be 3-5 sentences with no filler phrases. Compare against the benchmarks at https://github.com/hexiecs/talk-normal for expected output.

Pitfalls

  • Security scanner: Hermes flags this skill as "DANGEROUS" because it modifies 
    AGENTS.md
    (persistent prompt modification). This is expected — the skill's entire purpose is injecting always-on prompt rules. Install with 
    hermes skills install --force hexiecs/talk-normal/skill-hermes
    .
  • Rules only take effect in a new session — Hermes freezes context files at session start.
  • If you have both 
    .hermes.md
    and 
    AGENTS.md
    , the installer writes to whichever it finds first (
    .hermes.md
    > 
    HERMES.md
    > 
    AGENTS.md
    ).

Upstream and issues