claude-skills

A collection of Claude Code skills, cleaned up for public reuse. Each one is self-contained, dependency-light, and softcoded so it works on any project.

A skill is a folder with a SKILL.md (instructions Claude reads) plus optional helper scripts. Claude loads it automatically when your task matches the skill's description — you don't call it by name. See the skills docs.

Browsing with an AI? Point it at skills.json — a machine-readable index of every skill (summary, when to use / when to skip, deps, triggers, relationships) so it can recommend the right one. Humans: use the chooser below.

Which skill do I need?

They split into three themes — documents, collaboration (PRs), secrets — plus a machine-to-machine handoff. Two pairs are designed to face each other:

  COLLABORATION                       SECRETS
  review-contrib  ⇄  contribuer       autocli-password  ⇄  vault-secrets
  (maintainer)       (contributor)    (one interactive secret)  (a whole vault, replaces .env)

The skills

deck-builder

Build or restructure polished, on-brand PowerPoint (.pptx) decks programmatically. Encodes hard layout/design rules (fill-the-space, assertion titles, one-idea-per-slide, readable type floors, section-rhythm backgrounds), embeds true vector SVG icons, and self-verifies with silent audits + a render-to-PNG check before handing back. Theme-agnostic.

• Use it when you're creating a pitch/deck, or fixing slides that feel empty, repetitive, or inconsistent.
• Skip it when you only need to read/extract from a .pptx (the base pptx skill covers that).
• Needs Python (python-pptx, Pillow); rsvg-convert for icons; LibreOffice for the faithful render.
• README · SKILL.md

review-contrib

Audit, test and merge a collaborator's commit / branch / PR safely — before it touches main. A deterministic pipeline: discover the commit (fork/branch/PR), conflict-audit with merge-tree, review in an isolated git worktree, run the repo's validation gate, delegate a code review to a sub-agent, manually test, then merge only on approval.

• Use it when you're the maintainer adopting an incoming contribution and main is production.
• Skip it when you're the one writing the contribution → use contribuer instead.
• Needs git (2.38+) and the gh CLI.
• README · SKILL.md

contribuer

Prepare a clean, adoptable PR to a repo you don't maintain. The contributor-side mirror of review-contrib: sync the fork, branch off upstream/main, find and respect the project's immutable rules & conventions, pass its validation gate, and open a small reviewable PR — so whatever the reviewer checks, you did it first.

• Use it when you start coding a feature/fix on someone else's repo.
• Skip it when you're the maintainer reviewing others' work → use review-contrib.
• Needs git and the gh CLI.
• README · SKILL.md

autocli-password

Run a CLI command that needs a secret only the user knows — without pasting it. Claude runs the command; a native hidden-input dialog pops up; the value flows straight into the command's environment. The secret stays in RAM — never in the chat, shell history, or on disk. Optional ssh-agent-style RAM cache to avoid retyping.

• Use it when a single command blocks on a credential (decrypt a vault, an SSH passphrase, a sudo step).
• Skip it when you need many secrets at runtime / to replace .env → use vault-secrets.
• Needs Bash + Node (zero external deps). macOS/Linux dialogs, TTY fallback.
• README · SKILL.md

vault-secrets

Choose & configure a queryable secrets manager, and use it in place of .env. A provider-agnostic façade (vault.sh) over Bitwarden, 1Password, pass, KeePassXC, Doppler, Infisical: store a secret (masked input, never pasted), read one, or inject a whole set into a command at runtime — so you can delete the plaintext .env.

• Use it when you want secrets out of .env, or help picking/wiring a password manager into scripts/CI.
• Skip it when you just need one interactive secret for one command → use autocli-password.
• Needs Bash + one supported vault CLI (Node for the Bitwarden adapter; no jq).
• README · SKILL.md

handoff

Hand a Claude Code session between machines over git. Transports the code (a WIP commit on a safe branch) + a HANDOFF.md resume brief, and prints a ready-to-paste resume command. Refuses to push WIP onto an auto-deployed branch. Optional autonomous remote agent (delegate) and a config-driven target diagnostic (check). Softcoded per repo via .claude/handoff.json.

• Use it when you want to continue (or delegate) work on a VPS / another workstation.
• Skip it when both ends already share a live workspace.
• Needs Node (zero deps) + git/SSH on both machines.
• README · SKILL.md

Install

Skills live in ~/.claude/skills/ (global) or <project>/.claude/skills/ (per-project). Copy the one(s) you want:

git clone https://github.com/lbachelotcapitalb/leo-bachelot-ia-skills.git
cp -R leo-bachelot-ia-skills/skills/deck-builder ~/.claude/skills/

No registration step — Claude Code discovers them and triggers the right one from your task description. Some skills reference each other (e.g. handoff and vault-secrets can use autocli-password); installing the companion is optional, each degrades gracefully.

Notes

• Genericized. Extracted from a personal setup; any reference to private infra (vault names, hosts, project conventions) was replaced with neutral placeholders. No personal data remains.
• Language. Some skills are written in French, some in English — the description is what Claude matches on, and it operates fine in either language.
• Platform. Mostly cross-platform; a few use macOS-first native dialogs / the LibreOffice render path with documented Linux fallbacks. Each skill's README lists its exact prerequisites.

License

MIT. deck-builder reuses MIT-licensed design-token tables and text-measurement ideas — attribution is in skills/deck-builder/SKILL.md.