authorial-judgment
Skill6 starsChanged 52 days agoMITView on GitHub
Use when a draft a reader will judge the writer by needs an editorial pass: essays, posts, talks, newsletters, website and product copy, cover letters, recommendation and reference letters, outreach and customer emails; or when prose "sounds like AI", reads generic, or should sound like the author. NOT for README/API docs, specs, changelogs, release notes or code.
In this repository27 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- agent-safety-reviewSkill
- agent-team-executionSkill
- agent-team-reviewSkill
- alert-hygieneSkill
- batch-scriptingSkill
- capture-knowledgeSkill
- deploy-gateSkill
- design-debateSkill
- implementation-drift-checkSkill
- improvement-minerSkill
- incident-analysisSkill
- incident-trend-analyzerSkill
- openspec-shipSkill
- outcome-reviewSkill
- panelSkill
- product-discoverySkill
- project-verificationSkill
- prototype-labSkill
- runtime-validationSkill
- second-opinionSkill
- security-scannerSkill
- skill-scaffoldSkill
- supply-chain-investigationSkill
- synthesizeSkill
- unified-context-stackSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 8 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈8
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 29
If none of those exist, keep the prose clean.
- Distinctive: line 35
Read the draft once and decide which moves it needs.
- Distinctive: line 41
If the draft is a cover letter or application (addressed to a hiring manager, recruiter or committee, about the writer's fit): move 2 means …
- Distinctive: line 49
Authorial position, not persona: Write from what the author has actually seen, believes, is skeptical of, or refuses to overclaim — not an …
- Distinctive: line 55
Real objection = real friction: Add friction only when the draft leaves unanswered a pushback a smart reader would raise.
- Distinctive: line 76
AI-inversion refusal: Before finishing, name the default move a model would make here (the safe hook, the tidy symmetrical close, the …
- Distinctive: line 84
Skip the baseline; name the blind spot: Don't explain what the target reader already knows, and don't use false-authority cues ("as is …
- Distinctive: line 93
Before the style pass, ask: Is the argument doing real work?
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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
Authorial Judgment
A revision lens for prose that is technically correct, well-structured, and still dead. LLM prose tends to be too smooth, too complete, too linear, too certain, too eager to resolve every thought. This lens makes judgment visible instead of faking imperfection.
Use this as a post-draft pass — after the premise is settled, before the final style pass. Draft first (or take the user's draft), then run the six moves below.
When NOT to use:
- README, API docs, tutorials — reference writing should be complete and linear.
- Specs, changelogs, release notes, status updates: this lens is not their default editing pass. Precision about uncertainty still applies to them; the moves do not.
- Code, commits, tests, config.
The hard gate: real texture or clean prose
Human texture comes from exactly three sources:
- Real details the author supplied.
- Real uncertainty in the evidence.
- Real editorial judgment about what matters.
If none of those exist, keep the prose clean. Do not decorate the absence. Never fabricate uncertainty, lived experience, memory, cognitive bias, or knowledge gaps to sound human. This gate outranks every move below.
Which moves the draft needs
Read the draft once and decide which moves it needs. A move you cannot name a reason for is not applied; the author's request is a reason. A good draft comes back near-unchanged: the author's facts, numbers, dates, names, commitments and stated certainty stay as written, and a plain claim stays plain unless the draft's own evidence contradicts it. End with the three largest edits and why.
If the draft is a cover letter or application (addressed to a hiring manager, recruiter or committee, about the writer's fit): move 2 means answering a concern the posting states, with facts already in the letter. A gap the applicant mentions outside the letter goes in your note to them, not in the letter. The close asks for the conversation without qualifying the fit.
The six moves
- Authorial position, not persona. Write from what the author has actually seen, believes, is skeptical of, or refuses to overclaim — not an invented biography. Weak: "As someone who spent years in boardrooms…" Better: "The boardroom version is simple: everyone agrees in principle, then avoids the one decision that makes the strategy real."
- Real objection = real friction. Add friction only when the draft leaves unanswered a pushback a smart reader would raise; if the draft already answers it, move on. If answering it needs material the author has not supplied, name the objection in your note to them, not in the piece. Ask: what would they reject, think is ignored, or consider too easy? Then answer it briefly. Not random detours.
- Deliberation only when earned. Show the thought; don't announce it. Kill announced-thought phrases: "let's unpack," "it's worth noting," "as we delve deeper," "the deeper point is," "before we proceed." Use a visible turn only when the claim is complex, contested, or easy to oversimplify.
- Sharpening re-articulation. Revisit a key idea at most once, and only when the second pass adds a mechanism, a consequence, or a more precise noun. Restating the same idea in prettier words is repetition, not depth.
- Rhythm follows thought. Vary sentence and paragraph shape to match the argument, never mechanically. If several sentences start the same way or every claim is followed by an example in the same pattern, restructure — lead one sentence with the consequence, another with the constraint; break a long explanation before the reader has to work.
- AI-inversion refusal. Before finishing, name the default move a model would make here (the safe hook, the tidy symmetrical close, the reflexive metaphor, the paragraph included only to sound complete, the overstated claim). Write the alternative. Keep whichever of the two states a mechanism or a consequence the other does not. If both do, or neither does, keep the author's version.
Also
- Skip the baseline; name the blind spot. Don't explain what the target reader already knows, and don't use false-authority cues ("as is generally understood," "the discerning reader"). Name the actual blind spot.
- Earn the aside. A digression is allowed only if it makes the main point sharper — reveals a mechanism, exposes a tradeoff, or connects to a concrete operating reality.
- No decorative metaphor. An image must clarify the mechanism, not sound nice.
Final review
Before the style pass, ask: Is the argument doing real work? Is the smart reader's objection addressed? Is any uncertainty real rather than performed? Did any re-articulation deepen the idea? Is there any fake memory, persona, or flaw? Is every fact, number, date and stated certainty as the author wrote it? Could a smart reader replace this with a generic article and lose nothing? If yes, revise.
For failure-mode names and repairs, see references/red-flags.md.
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.

