agentsclimarketplace
Sign in

Slop update

Skill nadavc2c/SlopMachine/.claude/skills/slop-update

Monthly bleeding-edge refresh for SlopMachine — re-verify every library, dependency, model, template asset, and agent standard (Agent-Skill + Claude-Code-plugin) against the LATEST STABLE official docs, update the in-code PERIODIC-UPDATE-RESEARCH tags, and log it. Use when the user asks to "update everything", "refresh to latest", "monthly update", or "bring deps/models up to date".From its SKILL.md

Install
npx -y skills add nadavc2c/SlopMachine --skill slop-update

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

  • 2 stars2 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

3.0 KB, as published. Nobody here has run it

/slop-update — monthly bleeding-edge refresh

Keep SlopMachine on the latest STABLE, official libraries, models, and agent standards. Read the policy + hard constraints first: docs/PERIODIC-UPDATE.md. Core rule: verify everything against live official docs — never assert from memory — and use latest STABLE only (no beta/rc/preview).

Procedure

  1. Enumerate the tags. rg "PERIODIC-UPDATE-RESEARCH" across the repo; sort by oldest last-verified.
  2. Libraries / dependencies (pyproject.toml): for each pin, fetch the live PyPI JSON / official release notes; compare floor vs latest stable. torch especially: confirm the chosen CUDA index (currently cu130) still carries the latest cp314 Blackwell-sm_120 Windows wheel — PyTorch rotates indexes out (that is how 2.11→2.12 stranded us). Verify each upgrade still satisfies the target stack in docs/PERIODIC-UPDATE.md.
  3. Models (src/slopmachine/config/models.yaml): use the slop-models skill to re-verify every repo_id still exists, is ungated/loadable, is diffusers-native on the installed diffusers, is STABLE (not preview), and fits 16 GB; surface any newer current-best (e.g. a newer FLUX/Qwen/Wan, or real ungated weights for a model that was paper-only).
  4. Template / mascot media (docs/, any catalog assets, config/styles/*): verify present, license-clean, and still load.
  5. Agent-Skill standard: re-check .claude/skills/*/SKILL.md frontmatter + structure against the live agentskills.io specification.
  6. Claude-Code-plugin standard: re-check .claude-plugin/marketplace.json + plugin.json against the live code.claude.com plugin/marketplace docs (+ SchemaStore schemas); re-check llms.txt vs llmstxt.org.
  7. Apply + tag: make the clearly-current upgrades; bump each last-verified date + source URL; uv lock
    • uv sync.
  8. Verify on-device: uv run slop info (sm_120, ~16 GB) + a smoke uv run slop image (and uv run --extra dance slop dance if touched).
  9. Log it: append a dated entry to the "Revalidation log" in docs/PERIODIC-UPDATE.md — what was current, what was bumped, what is newly stale, and any item deliberately NOT adopted (paper-only / closed-API / blog-claimed) with the reason.

Ignore (do not chase)

Blog "open weights" claims with no HF repo; paper-only models with no diffusers-loadable weights; closed cloud-API-only releases. Adopt only real, ungated, diffusers-native, stable weights.

Keep looking

Skills are one crate of 627,801. 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.