Curate skills

Skill Chulinuwu/skillloom/skills/curate-skills

Private, local-first second brain for AI agents, with shared memory and governed skills across Claude Code, Codex, and your Tailnet.

Install

npx -y skills add Chulinuwu/skillloom --skill curate-skills

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

21 days oldThe repository was created 21 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
0 stars0 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.

What its author says it does

Copied from the file, not written here

Inspect and improve Skillloom-managed Agent Skills with base-hash protection, validation, trust scanning, explicit approval, promotion, and rollback evidence. Use when a managed skill is stale, duplicated, unclear, failing in real use, or needs a reviewed update. Do not use to silently rewrite installed skills or promote changes autonomously.

SKILL.md

2.3 KB, 400 tokens by cl100k_base, as published. Nobody here has run it

Curate Skills

Improve a managed skill from observed evidence while preserving its provenance and the user's ability to reject or reverse the change. Respond in the language of the user's latest message. Keep exact identifiers, hashes, paths, and source titles unchanged.

Workflow

Identify the installed skill, its active path, the behavior that failed or drifted, and the evidence supporting a change.
Inspect the full package and its promotion history. Separate confirmed defects from stylistic preferences.
Copy the skill to a temporary directory and make the smallest coherent revision. Keep SKILL.md harness-neutral, remove stale resources, and do not add credentials, transcript bodies, symlinks, binaries, or machine-specific tool instructions.
Run skillloom capture <revised-skill-directory> --created-by agent --base <installed-skill-path> --evidence <reference>. The recorded base hash protects concurrent human edits.
Run skillloom validate <candidate-id>. Compare the candidate with the installed base and review every trust finding. The scan is preflight evidence, not a malware-proof sandbox.
Present the behavioral change, diff summary, provenance, base and candidate hashes, findings, intended targets, and reload impact. Stop on danger findings and ask before accepting warnings or promoting.
Only after explicit approval, run skillloom promote <candidate-id> --target <targets> --scope <scope> --yes. Add --accept-warnings only for warnings the user explicitly accepts.
Verify destination hashes and the promotion record. If the base changed after capture, do not force an overwrite; recapture from the new base.
If verification fails, inspect skillloom status. Resume only when staged evidence still agrees. Otherwise run skillloom rollback <promotion-id> --yes and report the rollback record and restored hashes.

Never edit an active destination directly and never promote autonomously.