agentsclimarketplace

Grizzly draft

Skill HarishDvs/Grizzly/skills/grizzly-draft

Expand the author's rough material into prose. Use when the user gives a snippet, outline beat, or scene description and asks to expand it, write it up, flesh it out, or draft it. The highest-risk mode; runs on strict rails.From its SKILL.md

Install
npx -y skills add HarishDvs/Grizzly --skill grizzly-draft

Assembled 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

3.4 KB, as published. Nobody here has run it

Grizzly Draft — Drafting on Rails

This is where generated prose drifts into AI mush, so it runs on the strictest rails in the suite.

Scene readiness

Load decks/scene-engine.md before drafting. Form its compact scene contract from the author's material. If a load-bearing element is missing, ask one focused question rather than compensating with polished filler.

The rails

  1. Work in small units by default. Take the smallest coherent piece, hand it back, and wait for the author's reaction. If the author explicitly requests a full scene, draft it only after the scene contract and any meaning-changing choices are settled; still present it as a proposal.
  2. Stay inside the author's material. Expand what they gave you. Don't invent new plot, characters, or world elements to fill space. If the snippet is thin, the draft is short; ask what they want to add rather than padding it yourself.
  3. Anchor on the author's voice, not generic "good prose." Before drafting, re-read VOICE.md's anchor lines and a nearby passage of the actual manuscript, and write toward that. When in doubt, underwrite: a plain true line beats a decorated one.
  4. Genre pack applies. If NOVEL.md names a pack, its pacing and hook norms shape where the beat sits in the chapter, never how the sentences sound (that is VOICE.md's job alone).

Before showing anything (four passes, in order, every time)

This is the highest-risk mode in the suite. These passes are not advisory and not optional here: never present prose that has not passed tell-scan and an independent read. Self-review alone is unreliable because the model is blind to its own default cadence.

  1. Tell-scan (the HARD floor). Run the scanner on the draft: powershell.exe -ExecutionPolicy Bypass -File tools/tell-scan.ps1 <file>. When you have not written a file yet, mentally apply the same HARD rules: em dash, "too X to", "not X but Y", report-frames (realized / noticed / felt / sensed). Any unresolved HARD hit must be fixed before presenting. Author-blessed constructions follow VOICE.md; preserve lines are reported and left untouched. Tell-scan is the floor, not the ceiling.
  2. Fresh-reader blacklist audit. Re-read the lines against decks/blacklist.md and VOICE.md. Use an independent reader when the current harness and author permit it; otherwise perform a sequential second read. Catch the inserted simile, wistful relative clause, and fragment-then-lyrical cadence.
  3. Register and tense pass. Run decks/register-tense.md: the past-tense spine, the trailerese maxim ("We hold the line"), the present-tense leak in past-tense narration, modern-idiom and clinical leaks. A word-clean line can still read modern; the cure is grammar and cadence, never archaism.
  4. Delivery pass: run decks/delivery-pass.md. Name the landing of the beat you drafted. If you cannot, say so and bring the benchmark comparison rather than passing off flat reportage as finished prose.

Presentation

Present as a proposal, never a delivery: "Here's one way that beat could go," then the lines, then "keep going, tighten it, or try a different angle?" The author is choosing, not receiving.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most docs writing skills give

Counted across 1,951 of the 3,904 authors here whose files we hold, read 2026-09-06

  • Use third-person for skill descriptionsin 54 of 1951, across 35 files
  • Start descriptions with Use whenin 43 of 1951, across 29 files
  • Run baseline scenarios before writing any skillin 40 of 1951, across 26 files
  • Use active voicein 40 of 1951, across 36 files
  • Map file responsibilities before defining tasksin 36 of 1951, across 29 files
  • Use checkbox syntax for tracking stepsin 35 of 1951, across 27 files
  • Ask one question at a timein 35 of 1951
  • Offer execution options after saving the planin 33 of 1951, across 24 files
  • Include complete code in every stepin 33 of 1951, across 27 files
  • Design units with clear boundaries and interfacesin 31 of 1951, across 23 files
  • Announce the skill usage at the startin 30 of 1951
  • Verify agent compliance after adding the skillin 29 of 1951, across 17 files

Said here and by no other author read

  • Load the scene engine before drafting
  • Work in small units by default
  • Expand only the provided material
  • Do not invent new plot or characters
  • Anchor writing on the author's voice
  • Underwrite when in doubt

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once.

Keep looking

Skills are one crate of 613,373. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.