From claudius
Use when developing code. Universal rules for TDD, self-review, quality timing, review format, security. Preloaded on developers.
How this skill is triggered — by the user, by Claude, or both
Slash command
/claudius:coding-best-practicesThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Universal rules for all developer agents. Language-specific guidance lives in each agent's own instructions.
Universal rules for all developer agents. Language-specific guidance lives in each agent's own instructions.
Steps 3-5 of every developer workflow (after build environment and prior art check):
Only run formatting, linting, and tests right before committing (or when the user explicitly asks). Don't run them after every edit — it wastes time and tokens.
Never re-run a build, test, or lint command just to see more of its output. Capture full output on the first run using tee: f=$(mktemp /tmp/build-XXXXXX.txt) && <command> 2>&1 | tee "$f" | tail -80 && echo "Full output: $f". If the visible tail is insufficient, read the temp file — do not re-execute the command.
Use the report-format skill for output structure. IDs are provisional (consolidation reassigns them).
git blame. Acceptable exceptions are rare: citing an external constraint (upstream issue, RFC, kernel API quirk) that justifies a non-obvious current choice. Composes with rust-best-practices M-NO-TOMBSTONES — same principle, different framing.// comments, private/non-pub rustdoc, module headers that just rephrase the file name. The cap relaxes to 5–10 lines for public API rustdoc that genuinely teaches downstream callers — parameters, return values, preconditions, error semantics, panic conditions, one-line examples. Both tiers obey present-state. The relaxation is on length, not on history. (See rust-best-practices C-EXAMPLE, C-FAILURE, M-FIRST-DOC-SENTENCE, M-CANONICAL-DOCS.)CMT-001, SEC-014, CODE-007, RUST-123, PROJ-002, CALL-005, etc.) in source code, comments, READMEs, or any other committed file. These IDs are reassigned every time the consolidator runs and become dead references after merge. Allowed ID forms (permanent / standards-body / repo-permanent) include ADR-NNN, RFC-NNN, CWE-NNN, CVE-YYYY-NNNN, OWASP-A0N / OWASP-LLM0N / OWASP-API0N, ATT&CK IDs, GHSA-…, GitHub issue/PR refs (#1234, org/repo#1234), TODO / FIXME / XXX / HACK comments, and test-case IDs from a committed test-spec document. Rule of thumb: if the ID is born inside a regenerated JSON / triage report, it's forbidden in committed code; if the ID lives in a committed Markdown / YAML / standards-body doc, it's fine. Enforced advisory by scripts/lint_ephemeral_ids.py (reviewer-side) and by every developer agent that preloads this skill (write-side).search_standards MCP tool (if available) to check coding and security standards when facing unfamiliar patterns or compliance questions.Tests must never touch real user data. Override XDG_CONFIG_HOME/XDG_DATA_HOME/HOME/app-specific env vars to temp dirs. Use in-memory or temp-file DBs, mock external services, write only to tmp//mktemp paths, use fake credentials.
Rust: use the tracing crate (not log).
| Level | Use for |
|---|---|
error | Important / fatal errors — things that need attention |
warn | Less significant errors — degraded but recoverable |
info | Business events — user-visible actions, state transitions, milestones |
debug | Secondary execution paths — error handling branches, fallback logic |
trace | Primary path execution — normal flow, detailed step-by-step progress |
Never log inside hot loops or frequently called code paths — even at trace level. Log before/after the loop, or log a summary (count, duration) once it completes.
Before finishing, commit all changes with a descriptive message. Never leave uncommitted work. Never commit to main/master. Run git status to confirm clean state before exiting.
npx claudepluginhub lklimek/agents --plugin claudiusProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.