claudia
Health Warn
- License — License: MIT
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Low visibility — Only 6 GitHub stars
Code Warn
- network request — Outbound network request in eval/fixtures/api-client/src/lib/apiClient.ts
Permissions Pass
- Permissions — No dangerous permissions requested
No AI report is available for this listing yet.
Ponytail + Caveman + Clean Code
claudia
Minimalism persona for Claude Code: 7-rung YAGNI ladder + delegation ladder
- dry/deadpan voice + one-sentence completion rule. Installs as user-level
files under~/.claude/— applies to every project, no per-repo setup.
Personal tool, one calibrated mode, Claude Code only.
Components
| File | Job |
|---|---|
CLAUDE.md.snippet |
always-on: minimalism + delegation ladders, safety floor, completion rule, ceremony suppression. Appended to ~/.claude/CLAUDE.md. |
output-styles/claudia.md |
voice layer: dry/deadpan, structured output. Select via /config. |
skills/fresh-work |
plan+Q&A pass, self-triggers on new work; defers to plan mode. |
skills/claudia-debt |
harvests // claudia: debt markers; on request or unprompted after a new marker. |
skills/doc-router |
splits a bloated CLAUDE.md into router + agent reference + human docs; self-triggers on bloat. |
agents/claudia.md |
opt-in subagent: voice + ladders for delegated code-writing. |
agents/bulk-reader.md |
haiku-pinned read-only summarizer: answers questions about large files without pulling them into the main context. |
hooks/read-guard.sh |
PreToolUse guard: denies whole-file Read (and cat/head/tail/less/more) past CLAUDIA_READ_GUARD_LINES (default 350; 0 off), redirecting to bulk-reader. Registered in settings.json at install. |
hooks/session-boot.sh |
SessionStart hook: injects a short claudia pre-flight reminder (delegation + skill self-triggering) into every new session. CLAUDIA_SESSION_BOOT=0 disables it. Registered in settings.json at install. |
Install
git clone https://github.com/joe-signorile/claudia.git
cd claudia
./install.sh
/config → Output style → claudia, or skip the prompt:
./install.sh --set-output-style
Remove:
./uninstall.sh
Symmetric, re-runnable, backs up a pre-existing conflicting file to .bak
once.
Scope
Benchmarked in eval/ — self-judged, directional. No
multi-host adapters, no telemetry. Voice reaches the main thread and the
opt-in claudia agent, not arbitrary subagents; the ladders still reach
code-writing subagents via user-level CLAUDE.md.
Scope change: claudia previously shipped no hooks. It now installs two,
both registered in settings.json, active in every project. ThePreToolUse read guard is real enforcement — the delegation ladder alone
is guidance the model can decline, the guard is not. The SessionStart
hook is not: it injects a short pre-flight reminder at the start of every
session so the ladder/skill-triggering norms stay fresh and prominent
independent of where they land in CLAUDE.md, but it can't force a
delegation decision — no hook can, short of something byte-checkable like
the read guard's line count. Both fail open on anything they can't
evaluate; CLAUDIA_READ_GUARD_LINES=0 and CLAUDIA_SESSION_BOOT=0 turn
each off independently. ./uninstall.sh strips both registrations and
leaves any hooks claudia didn't install alone.
Repo work: start at agents.md.
MIT — LICENSE.
Reviews (0)
Sign in to leave a review.
Leave a reviewNo results found