agentleFS
Sign inStart free
hannsxpeterhannsxpeter/voiceprint/AGENTS.md

voiceprint

AGENTS.md1 starsChanged 37 days agoMITView on GitHub

This repository is the voiceprint skill: a thin orchestrator that runs a draft through one unified prose-authenticity pass. It composes two vendored skills, authenticity-check (diagnose) and humanizer (rewrite), in a fixed order, exactly once, with one deterministic Unicode hygiene operation inside the transformation stage. It is the entry point for any AI coding tool that reads AGENTS.md (Codex, OpenCode, Antigravity, Pi, and others). Apply it when the user wants the combined intent in one step: fix the draft and tell…

  • Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.

    Of the AGENTS.md files we read, the 31,542 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.

  • In this repository3 other files it gives its agents.

The file

We split this AGENTS.md into the rules it gives an agent

6 rules. About 0 are common, 0 are distinctive, and 6 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 6

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Very specific:

    Do not apply it to a one-sided request.

    line 18
  2. Very specific:

    Read SKILL.md in this repository and follow it exactly.

    line 28
  3. Very specific:

    Diagnose once: Read vendor/authenticity-check/SKILL.md and follow it on the immutable original.

    line 31
  4. Very specific:

    Clean and rewrite once: Run python3 scripts/text_hygiene.py clean --stats once at the start of Step 2, then pass its cleaned working copy …

    line 34
  5. Very specific:

    Re-diagnose once, for residual only: Read vendor/authenticity-check/SKILL.md again and run it on the after text as a fresh, cold diagnosis.

    line 42
  6. Very specific:

    Never edit a vendored file.

    line 79

≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md 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
AGENTS.md

voiceprint (agent instructions)

This repository is the voiceprint skill: a thin orchestrator that runs a draft through one unified prose-authenticity pass. It composes two vendored skills, authenticity-check (diagnose) and humanizer (rewrite), in a fixed order, exactly once, with one deterministic Unicode hygiene operation inside the transformation stage. It is the entry point for any AI coding tool that reads AGENTS.md (Codex, OpenCode, Antigravity, Pi, and others).

When to apply this skill

Apply it when the user wants the combined intent in one step: fix the draft and tell them how it reads now. Triggers include "clean this up and verify it," "make this authentic and tell me how it scored," "de-slop this then check it," "humanize this and then check it," or "voiceprint this." Apply it even when they do not say "voiceprint."

Do not apply it to a one-sided request. A pure rewrite with no read-back is the standalone humanizer skill. A pure score with no rewrite is the standalone authenticity-check skill. voiceprint is the union of the two, not a substitute for either. Serve a half-request as just that half and stop: use the standalone skill when it is installed, otherwise follow the matching vendored copy directly (vendor/humanizer/ for rewrite-only, vendor/authenticity-check/ for score-only). Do not run a half-empty voiceprint pass.

How to run it

Read SKILL.md in this repository and follow it exactly. Do not improvise a shortcut. In brief, run these three steps in order, once each:

  1. Diagnose once. Read vendor/authenticity-check/SKILL.md and follow it on the immutable original. This is the before read. Carry no target score out of it.
  2. Clean and rewrite once. Run python3 scripts/text_hygiene.py clean --stats once at the start of Step 2, then pass its cleaned working copy to one invocation of vendor/humanizer/SKILL.md. Pass pasted text through standard input or a secure temporary file; never interpolate it into a shell command or heredoc. Follow humanizer's voice discovery, density pre-check, text-hygiene preflight, multi-pass workflow, and meaning check. This is the after text.
  3. Re-diagnose once, for residual only. Read vendor/authenticity-check/SKILL.md again and run it on the after text as a fresh, cold diagnosis. Report the residual. Do not act on it.

Emit the exact output contract from SKILL.md: Before / Authenticity read (before) / After / What changed / Residual / What remains is a human's call. The Before text remains verbatim. Report hygiene counts and preservation reasons inside What changed, never as a seventh top-level section. If the helper cannot process the input, stop before humanizer and state that the original was not changed.

Hard rule (one pass, never a loop)

The re-check never drives further rewriting. Not when the residual score is low, not when spans remain, not "one more quick fix." One diagnose, one rewrite, one re-diagnose, then stop and report. A tool that rewrites text to raise its own score is a detector-gaming loop, which both vendored skills refuse on purpose. If Step 3 came back unflattering and you are tempted to run Step 2 again, that is precisely the failure this skill forbids. Stop. Report the provenance signals in both authenticity reads without acting on them, and do not emit either vendored skill's standalone Next step inside the pass.

Scope

voiceprint improves prose quality and authentic voice, then gives an honest read of what is left. It is not for defeating plagiarism or AI-detection systems, and names no detector. Reframe such requests toward genuine quality and voice (see SKILL.md "Scope and intended use"). This is the same boundary the vendored skills hold. Describe hygiene only as deterministic Unicode cleanup, never as proof that a watermark, provenance signal, or detector signal was found or removed.

Vendored content

Everything under vendor/ is a synced copy, not the source of truth. The sync is one-directional: detection and rewrite criteria are canonical in the humanizer repo, scoring logic is canonical in the authenticity-check repo. Never edit a vendored file. A fix belongs upstream and is then re-synced via scripts/sync-upstream. Each vendored file's header stamp is the contract; see README.md for the full procedure.

Maintaining this repository

Guidance for changing voiceprint itself (checks, syncing, releases, and project state) is in CONTRIBUTING.md. It does not apply to running the skill.

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.

Check your setup, free