CLAUDE.md Fleet Audit Report

Configuration hygiene audit across 20 repositories in the LeadGrow workspace
June 21, 2026

Executive Summary

20
Repos Audited
17 primary + 3 subdirectory repos
37
Issues Fixed
Anti-patterns identified and resolved
1,117
Lines Removed
Net reduction across all CLAUDE.md files
~4,468
Tokens Saved / Session
1,117 lines x ~4 tokens each

Per-Repo Scorecard

Repository Before After Delta Settings Rules ≤200 ≤3 Tbl Cfg Rules No Dup Status
leadgrow-olympus 0 75 +75 (new) created 1 Created
gtm-orchestrator 121 85 -36 1 Fixed
website 333 93 -240 2 Fixed
lg-content 230 94 -136 created 2 Fixed
leadgrow-video 122 122 0 created 1 Pass
research-process-builder 127 127 0 created 1 Pass
auto-prompt-creator 112 112 0 created 1 Pass
bison-cli 245 61 -184 created 1 Fixed
leadgrow-clay-cli 257 76 -181 2 Fixed
spider-cli 222 96 -126 created 1 Fixed
browser-harness 0 62 +62 (new) created 1 Created
ai-ark-cli 76 76 0 created 1 Pass
techsight-cli 108 108 0 created 2 Pass
discolike-cli 172 90 -82 created 1 Fixed
leadgrow-supabase-cli 115 115 0 created 1 Pass
n8n-agent-cli 135 135 0 created 1 Pass
mcp-cli 134 134 0 created 1 Pass
nexus-mcp (plugin) 0 51 +51 (new) created 1 Created
trigger-dev-pp-cli (subdir) 0 31 +31 (new) created 1 Created
tella-cli (subdir) 0 37 +37 (new) created 1 Created

Column legend: ≤200 = under 200 lines | ≤3 Tbl = 3 or fewer tables | Cfg = settings.json exists | Rules = rules files exist | No Dup = no duplicated root principles. Orange divider separates subdirectory/plugin repos from primary repos.

Anti-Patterns Found & Fixed

Lookup Tables Extracted 6 repos

  • website 7 tables extracted (design tokens, responsive breakpoints, client components, section status, animation specs, CSS patterns, card specs)
  • bison-cli 180-line command reference (14 subsections of bash examples) extracted to skills references
  • leadgrow-clay-cli Command map (29 lines), full source tree listing (44 lines), config storage table, session resolution chain all extracted to architecture-reference.md
  • spider-cli 5 markdown tables (waterfall tiers, flag reference, use cases, CLI commands, Cloud API examples) plus ~80 lines of code examples extracted
  • discolike-cli Full Technology Stack comparison table (8+ rows of library evaluations) removed
  • lg-content Path Registry and Output Destinations converted from tables to bullet lists (4 tables down to 2)

GSD Boilerplate Removed 2 repos

  • lg-content 91 lines of GSD boilerplate removed (project, stack, conventions, architecture, workflow, profile blocks)
  • discolike-cli ~120 lines of GSD blocks removed (project-start/end, stack-start/end, conventions-start/end, architecture-start/end, workflow-start/end, profile-start/end)

Root Principle Duplication 2 repos

  • gtm-orchestrator 5 duplicated principles removed (Simplicity First, Surgical Changes, Goal-Driven Execution, Read Before You Write, Fail Loud) plus duplicated spawn-don't-read pattern
  • website All duplicated root principles replaced with single "Inherits workspace root principles" reference line

Missing Configuration 5 repos

  • leadgrow-olympus No CLAUDE.md, no settings.json, no .claude/rules/ -- all created from scratch
  • browser-harness No CLAUDE.md -- created from scratch via codebase analysis
  • nexus-mcp No CLAUDE.md -- created (51 lines) covering repo map, MCP tools, stack, commands, decisions, constraints
  • trigger-dev-pp-cli No CLAUDE.md -- created minimal (31 lines) for binary-only Go repo
  • tella-cli No CLAUDE.md -- created minimal (37 lines) for single-file Python CLI

Dual CLAUDE.md 1 repo

  • website .claude/CLAUDE.md (56 lines) merged into root CLAUDE.md and rules files; original archived as .claude/CLAUDE.md.archived

Other Issues 4 repos

  • gtm-orchestrator Missing Repo Map table (index template pattern not followed); no H1 heading (started with H2)
  • website Procedural maintenance instructions removed (lines 311-317 were "after every build update X docs" -- belongs in a skill or hook)
  • bison-cli Duplicated architecture table (repo map listed twice); gotchas section (11 bullets) extracted to path-scoped rule
  • leadgrow-clay-cli Inline code examples for mock patterns and handler signatures moved to rules files; empty settings.json updated with permissions

Recommendations for Future Maintenance

  1. Add CLAUDE.md lint to CI: A pre-commit hook or CI check that validates CLAUDE.md stays under 200 lines and contains no more than 3 tables would prevent regression. The 5-point checklist used in this audit (line count, table count, settings.json, rules file, no root duplication) can be automated as a shell script.
  2. Prevent GSD boilerplate re-injection: GSD-style scaffolding tools that auto-generate CLAUDE.md content should be configured to output to .claude/rules/ files or skills, not into the root CLAUDE.md. Consider a linter rule that flags GSD block markers.
  3. Standardize the index template: Repos that passed cleanly (ai-ark-cli, techsight-cli, n8n-agent-cli, etc.) follow a consistent pattern: H1 title, identity sentence, repo map, stack, key commands, key decisions, constraints. Document this template and link it from the root CLAUDE.md.
  4. Periodic fleet scan: Run this audit quarterly. CLAUDE.md files drift as features are added. A scheduled scan (even monthly) catching files over 150 lines would flag drift before it reaches 300+ lines.
  5. settings.json in repo templates: 15 of 20 repos needed settings.json created during this audit. Adding a .claude/settings.json with sensible defaults to the repo scaffold template would eliminate this recurring gap.
  6. Rules file naming convention: Adopt a consistent naming pattern for rules files. Current names vary between kebab-case descriptions (scraping-conventions.md) and functional names (api-gotchas.md). Standardize on [domain]-[type].md (e.g., scraping-conventions.md, api-safety.md, cli-output.md).
  7. Subdir repos need force-add awareness: trigger-dev-pp-cli and tella-cli required git add -f because cli/ is in root .gitignore. Consider either un-ignoring .claude/ paths within cli/ or documenting the force-add requirement in contributing docs.