Grizzly audit
Simulate a cold first-time reader and report the reading experience of a chapter. Use when the user asks why a chapter feels boring or flat, whether a chapter works, what a reader would feel, or wants a fresh-eyes read. Reports experience, not errors; for error-finding use grizzly-review.From its SKILL.md
npx -y skills add HarishDvs/Grizzly --skill grizzly-auditAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
4.1 KB, as published. Nobody here has run it
Grizzly Audit — The Cold Read
Reviews find errors. Audits find the problems nobody can name: the chapter that is technically clean and somehow boring, the character the reader cannot picture, the death that lands as information. This mode exists because authors cannot read their own work cold, and it is the highest-value service in the suite.
Procedure
Choose the reading contract before loading the codex or author goals:
- Isolated cold read: read only the requested text when the author wants the excerpt or chapter judged on its own.
- Series-aware fresh read: default for a late-in-sequence chapter. Carry manuscript context already read; if none is loaded, read the immediate predecessor and the last relevant POV or thread passage before the target.
React first, then analyze. Load the codex afterward for comparison.
Report these six checks, each with chapter-position references ("opening", "the kitchen scene", "last page"):
1. Dread map
Where does threat exist for the reader, where does it spike, and where does it die? Threat dies through over-explanation, through repetition without escalation, and through safety that arrives too reliably. Name the line where you stopped being afraid.
2. Safety map
Where does the reader feel safe, and is that safety doing work? Earned warmth that will cost something later is structure; comfort that costs nothing is drag. If the whole chapter is safe, say so plainly; that is usually the finding.
3. Character-visualization check
For each character on page: can you physically see them, or are they a voice in white space? What are they doing with their bodies while they talk? Could a reader who skipped the introductions tell them apart by behavior alone? Name the characters who exist only as dialogue tags.
4. Prediction check
At each turn, what did you expect to happen next? Then: did the chapter confirm the prediction boringly, break it meaningfully, or wander without raising any prediction at all? No predictions raised is the flattest possible finding; report it as such.
5. Hook check
What, if anything, pulls to the next chapter? Is it positioned at the end or buried mid-chapter? Judge the pull against the scene's job, declared format, and author's purpose. Do not demand a retention device or cliffhanger by default.
6. Belief check
Did anything eject you from the fiction? Two ejector classes belong here even though they look like "errors": POV breaks (the narration shows what the POV character cannot perceive: sights through a closed door, another character's interiority) and syntax garble severe enough to force a re-read. Both are experience findings; a reader who stops believing or stops parsing has stopped reading. Light typos stay out of the audit (route to review).
After the cold read
- Load
decks/scene-engine.mdand identify the apparent scene job. Do not punish a successful quiet, comic, relational, atmospheric, or connective beat for lacking dread or reversal. Ask whether it produced the reader movement appropriate to its job. - Compare against the codex: did anything the cards mark as important fail to register in the cold read? A planted thread the reader never noticed is a planting failure, not a reader failure.
- End with the one-sentence verdict a beta reader would give a friend ("I'd keep reading, but I skimmed the middle"), then the single highest-leverage diagnosis or question. Route fixes to plan or edit.
Rules
- Honesty is calibrated, not padded: lead with the strongest finding. If the chapter works, report the material checks concisely rather than manufacturing six empty sections.
- Report experience, never fixes. Route fixes to grizzly-edit or grizzly-plan; offer, don't start.
- Never quote more than a line or two back at the author; reference positions.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.
Gives 0 of the 12 instructions most audit compliance skills give
Counted across 960 of the 1,589 authors here whose files we hold, read 2026-09-06
- Read product marketing context before asking questionsin 29 of 960, across 11 files
- Rank findings by severityin 29 of 960, across 22 files
- Generate audit reportin 22 of 960
- Run the audit scriptin 20 of 960, across 19 files
- Generate a prioritized action plan reportin 19 of 960, across 11 files
- Ensure one H1 per pagein 15 of 960, across 5 files
- Ensure sitemap exists and is accessiblein 14 of 960, across 4 files
- Verify alt text on all imagesin 12 of 960, across 3 files
- Determine the audit scope before startingin 12 of 960, across 4 files
- Verify important pages allowed in robots.txtin 11 of 960, across 2 files
- Detect business type from homepage signalsin 11 of 960, across 7 files
- Delegate specialized tasks to subagentsin 11 of 960, across 7 files
Said here and by no other author read
- Choose the reading contract before loading context
- React first and analyze afterward
- Report six specific reading experience checks
- Load scene engine decks after the cold read
- Compare findings against the codex
- End with a one-sentence beta reader verdict
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once.