agentleFS
joelbarmettlerUZHjoelbarmettlerUZH/ultrainit.sh/prompts/CLAUDE.md

ultrainit.sh / prompts

CLAUDE.md8 starsChanged 7 months agoApache-2.0View on GitHub

14 markdown files, each the system prompt for one claude -p agent. Files are paired 1:1 with schemas in schemas/. The prompt defines task intent; the schema enforces output types. Every gather agent prompt requires file citations for every claim. No assertion without a real file path. This is enforced by explicit instructions: patterns.md includes "Do NOT invent patterns. Only report what you can prove exists." Any new gather prompt must include a similar evidence requirement. Both synthesizer.md and synthesizer-tooling.md…

The file

We split this CLAUDE.md into the rules it gives an agent

1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Every "Do NOT do X" must be paired with "Do Y instead." synthesizer.md explicitly forbids lone prohibitions.

    line 36

≈ “Common” is an estimate: at least 1 in 5 of the 28,029 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

prompts/ — Agent System Prompts

14 markdown files, each the system prompt for one claude -p agent. Files are paired 1:1 with schemas in schemas/. The prompt defines task intent; the schema enforces output types.

File Index

FilePhaseModelPurpose
identity.md1haikuProject name, languages, frameworks, monorepo detection
commands.md1haikuBuild, test, lint, format commands from all config files
git-forensics.md1sonnetHotspots, temporal coupling, bug density, ownership
patterns.md1sonnetArchitectural patterns, error handling, naming conventions
tooling.md1haikuLinters, formatters, pre-commit hooks
docs-scanner.md1haikuExisting docs inventory and coverage gaps
security-scan.md1haikuFiles needing protection, safe alternatives
structure-scout.md1sonnetDirectory map with roles and priority tiers
module-analyzer.md1 (Stage 2)sonnetDeep-dive per directory (spawned once per important dir)
domain-researcher.md3sonnetBusiness domain terminology and framework best practices
mcp-discoverer.md3sonnetMCP server recommendations from registry + GitHub
synthesizer-docs.md4 Pass 1sonnet[1m]Generate CLAUDE.md files from all findings
synthesizer-tooling.md4 Pass 2sonnet[1m]Generate skills, hooks, subagents, MCP config
synthesizer.mdreference—Master/reference prompt covering both passes (not used directly in pipeline)

Critical Rules

Evidence Mandatory

Every gather agent prompt requires file citations for every claim. No assertion without a real file path. This is enforced by explicit instructions: patterns.md includes "Do NOT invent patterns. Only report what you can prove exists." Any new gather prompt must include a similar evidence requirement.

Portability Test (Synthesizers)

Both synthesizer.md and synthesizer-tooling.md end with a checklist that includes the Portability Test: "Any skill that could be dropped unchanged into an unrelated project has failed." Every skill must contain ≥3 codebase-specific file references to pass. This is the most important quality gate in the entire system.

Prohibition + Alternative Pairing

Every "Do NOT do X" must be paired with "Do Y instead." synthesizer.md explicitly forbids lone prohibitions. This applies to CLAUDE.md content, skills, and subagent instructions. Do not add prohibitions to synthesizer outputs without alternatives.

Generic Phrase Ban

These phrases are explicitly banned in synthesizer.md and must never appear in generated CLAUDE.md or skill content: "best practice", "clean code", "SOLID principles", "maintainable", "readable", "scalable", "well-structured", "production-ready", "industry standard". Delete on sight.

YAML Angle-Bracket Prohibition

synthesizer-tooling.md explicitly forbids < and > in YAML frontmatter fields. This prohibition is absent from synthesizer.md, even though it also generates skill YAML. When editing synthesizer.md, add: "NEVER use angle brackets (< or >) in YAML frontmatter or description fields — they break YAML parsing."

Schema Coupling

Prompts describe output field names in prose; schemas enforce types. The two must stay in sync manually — there is no automated cross-validation. When modifying a prompt's output section, always update the paired schema in schemas/ in the same commit.

module-analyzer.md receives the directory path as context injected into the prompt string at call time in lib/gather.sh, not via a system prompt variable. If refactoring the invocation, the path must be interpolated before the claude -p call.

Complexity vs Phase

Gather agents: 13–55 lines (narrow, single output type, cost-sensitive — many run in parallel). Synthesizers: 101–295 lines (multi-artifact, quality checklist, embedded template structures). This ratio is intentional. Never add quality checklists to gather agents.

Known Issue: Three Overlapping Synthesizer Prompts

synthesizer.md is the comprehensive reference. The actual pipeline uses synthesizer-docs.md (Pass 1) then synthesizer-tooling.md (Pass 2). Using synthesizer.md as a drop-in replacement for both passes would skip the Pass 1→Pass 2 dependency (Pass 2 needs Pass 1's CLAUDE.md as source of truth). Do not conflate them.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.