literary-critic
AGENTS.md0 starsChanged 27 days agoMITView on GitHub
This file is the project contract for every assistant (Claude, ChatGPT, Gemini, others). CLAUDE.md imports it; other chatbots should be pointed at it first. Generic tooling that lets an LLM read a full novel manuscript, index it so it can move through it without filling its context, analyse it against a catalogue of checks, and hand the writer concrete, prioritised revision actions — then re-index each new draft and track what changed. Sessions mix two conversations: - Plain messages are…
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 32,671 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 repository22 other files it gives its agents.
- claude-arsenal/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- coverage-gapsSkill
- dep-upgradeSkill
- designSkill
- executionSkill
- explore-ideaSkill
- gate-checkSkill
- githubSkill
- initSkill
- litcriticSkill
- mutmut-reportSkill
- pin-checkSkill
- pypi-releaseSkill
- python-bootstrapSkill
- queue-addSkill
- queue-nextSkill
- queue-statusSkill
- reviewSkill
- session-endSkill
- shipSkill
- specifySkill
The file
We split this AGENTS.md into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 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.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 22
Don't wait for [[...]] to improve the tool: when something learned on a working novel is generic (it would help with other novels), turn it …
- Distinctive: line 56
Every location: chapter/section, approximate page when known, and the quoted first and last words of the paragraph.
- Very specific: line 62
Measure: Every capability ships with a metric and a gate (see status/plan.md).
≈ “Common” is an estimate: at least 1 in 5 of the 41,581 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.
Files
literary-critic — rules for any LLM working here
This file is the project contract for every assistant (Claude, ChatGPT, Gemini,
others). CLAUDE.md imports it; other chatbots should be pointed at it first.
What this repo is
Generic tooling that lets an LLM read a full novel manuscript, index it so it can move through it without filling its context, analyse it against a catalogue of checks, and hand the writer concrete, prioritised revision actions — then re-index each new draft and track what changed.
The two channels
Sessions mix two conversations:
- Plain messages are about the novel being worked on (the working novel). Answer in the language the writer uses or the novel is written in.
[[...]]marks talk about this repo (the tool). Repo artifacts are written in English.
Don't wait for [[...]] to improve the tool: when something learned on a
working novel is generic (it would help with other novels), turn it into a
generic change to the repo — and phrase it without any trace of the novel.
Hard rules
- Never add anything about specific working novels to the repo. No
manuscript text, titles, character or place names, plot details, findings,
indexes, letters or notes. Examples in docs and tests use only the
public-domain texts under
corpus/or invented placeholders. - Working files live in a workspace outside this repo (see
docs/workspace.md).litcriticrefuses a workspace inside a git work tree of this repo. - Run the content guard before every commit (
make guard; the pre-commit hook does it). It blocks manuscript file types outsidecorpus/and any term listed in a workspace'sguard-terms.txt. When you start on a new novel, fill itsguard-terms.txtwith title, character and place names. - Never modify the writer's manuscript (Word, Google Docs, any file) without explicit permission for that specific change. Default to comments or suggestions; direct edits only after the writer confirms.
- The writer writes. This tool diagnoses and explains — what, where, why,
which effect to aim for, which options exist. By default it never writes
prose for the novel. Proposed wording is a per-writer setting
(
prose_suggestionsin the profile: off by default,minorfor small fixes, on for suggested passages), asked at setup as "who writes?" with small fixes as the recommended answer: the exact text for a comma, a synonym, an agreement, each one approved by the writer, while every new sentence and passage stays the writer's. Help with wording first shows the AI-text notice (authorship and copyright, detection and watermarks, disclosure to publishers, the writer's voice; we recommend the writing stays the writer's, small fixes at most) and records the acknowledgement. Suggested passages are the writer's choice. Edits reach the manuscript only as text the writer approved. - Talk to writers plainly. Assume no technical skills (not even search). Every location: chapter/section, approximate page when known, and the quoted first and last words of the paragraph. One change per message while the writer is applying changes; clear, direct, patient.
- Script first. Anything deterministic (parsing, counting, diffing, indexing, validating, rendering reports) is a script with tests. The LLM is used for judgment: reading, interpreting, critiquing, writing feedback.
- Measure. Every capability ships with a metric and a gate (see
status/plan.md); evidence goes tostatus/evidence/. - Runtime code is Python standard library plus doc-control (
src/litcritic/), so the kit runs inside chat code interpreters. doc-control (itself stdlib only) does every generic document read and write; onlydc.pyimports it, and literary-critic never rebuilds what doc-control owns. Dev tools (ruff, mypy, pytest) are fine.
Layout
| Path | Purpose |
src/litcritic/ | The litcritic package and lc CLI (stdlib plus doc-control) |
checklist/ | Generic analysis-check catalogue (part 2) |
schemas/ | JSON schemas for the profile and the reader report |
prompts/ | Portable LLM prompts (usable from any chatbot) |
corpus/ | Public-domain texts for tests and examples |
docs/ | Architecture, workspace format, research |
status/ | Specification, plan, gate evidence |
tests/ | Pytest suite |
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.

