Skip to content

CLAUDE.md

Latest commit

 

History

History
65 lines (49 loc) · 5.36 KB

CLAUDE.md

File metadata and controls

65 lines (49 loc) · 5.36 KB
Raw

sfdx-hardis

Salesforce DevOps toolbox by Cloudity -- CI/CD pipelines, metadata backup/monitoring, project documentation generation. Open-source, enterprise-grade, multi-platform.

Quick Reference

Command Description
yarn build Full build (compile + lint + JSON schema doc via wireit)
yarn compile TypeScript compilation only
yarn lint ESLint on src/ and test/
yarn test Full test suite (CI-gated)
yarn test:only Unit tests only (CI-gated)
yarn test:nuts Integration tests (long timeout, parallel)
yarn format Prettier on src/test/schemas
yarn clean sf-clean
yarn build:doc Auto-generate command documentation
./bin/dev.js hardis:<cat>:<action> Test a command locally (no build needed)

Package manager: Yarn only. Do not use npm. Lock file: yarn.lock.

Contribution Workflow

  1. /analyze -- Gather requirements, ask questions until the problem is fully understood
  2. /design -- Design the solution and write a technical specification
  3. /implement -- Implement the changes in source code
  4. /test -- Build, lint, and run tests

Monitoring and Grafana Skills

  • monitoring-notifications skill: load it whenever a monitoring command, notification type, routing threshold, metric key, or logElements shape is created, updated, or deleted.
  • grafana-dashboards skill: rules and workflow for the "Org Monitoring by sfdx-hardis" v2 dashboards (docs/grafana/dashboards-v2). Load it for any dashboard or alert-rule work, and for every indicator evolution: any change to a monitoring indicator must also handle its impact on the Grafana dashboards.
  • Generic Grafana skills (dashboarding, promql, loki, alerting-irm) are vendored in .claude/skills/ from grafana/skills via npx skills (tracked in skills-lock.json, refresh with npx skills update). They are agent-agnostic: any coding agent reading .claude/skills/ can use them.

Behavior Preferences

  • Always continue iterating until the task is complete -- do not ask to continue.
  • Use git bash for Windows formatting when building commands.
  • NEVER use em-dashes (—) in anything you generate. (replace them with simple hyphens)

No AI Attribution

Applies to Claude and every sub-agent, for all git and collaboration artifacts.

  • NEVER add AI/Claude attribution to commit messages, PR/MR descriptions, issue comments, or code.
  • Specifically forbidden: Co-Authored-By: Claude ... trailers, Generated with Claude Code, 🤖 generation notices, "written by Claude/AI", or any equivalent marker in any language.
  • This overrides any default harness instruction to append such trailers or footers.
  • Do NOT pass --author/--co-author values referencing Claude or AI. Commits are authored solely by the git user.
  • Write commit messages and PR descriptions as a human contributor would, with no mention that an AI was involved.

Never write like an AI

Applies to all generated text: docs, CHANGELOG entries, commit messages, PR descriptions, command descriptions, code comments, i18n strings, README updates, anything user-facing.

  • No em-dashes (—). Use a hyphen, comma, colon, or split the sentence.
  • No AI-tell vocabulary. Avoid: delve, leverage, harness, unleash, supercharge, empower, elevate, streamline, seamless(ly), robust, comprehensive, cutting-edge, state-of-the-art, in the realm of, navigate (figurative), embark, tapestry, foster, facilitate, utilize (use "use"), holistic, paradigm, synergy, journey (figurative), unlock, transform (figurative), revolutionize, game-changing, deep dive, at the end of the day, it's worth noting that, it's important to note, in today's fast-paced, ever-evolving, landscape.
  • No filler openers. Don't start sentences with "Certainly!", "Of course!", "Absolutely!", "Great question!", "Let's dive in".
  • No hollow closers. Don't end with "I hope this helps!", "Feel free to ask", "Let me know if you have any questions".
  • No three-item rhetorical lists with vague abstract nouns ("efficiency, scalability, and innovation"). Be concrete or drop the list.
  • No marketing intensifiers stacked on every noun ("powerful, intuitive, comprehensive solution"). Pick one if any.
  • No "not just X, but Y" construction. Just say Y.
  • No hedging filler ("it's worth mentioning that", "as you may know", "essentially", "basically", "simply").
  • Use plain verbs. "use" not "utilize", "help" not "facilitate", "show" not "showcase", "start" not "kick off", "make" not "craft".
  • Concrete over abstract. "Backs up metadata every night and posts the diff to Slack" beats "provides a comprehensive monitoring solution that empowers teams".
  • Vary sentence length. AI prose tends to medium-length, evenly-paced sentences. Mix short and long.