Commit graph

9 commits

Author SHA1 Message Date
Florian BRUNIAUX
8a4d116e2e feat(docs): add LLM Handbook + Google Whitepaper integration v3.3.0
Advanced Guardrails:
- prompt-injection-detector.sh (PreToolUse)
- output-validator.sh (PostToolUse heuristics)
- claudemd-scanner.sh (SessionStart injection detection)
- output-secrets-scanner.sh (PostToolUse secrets leak prevention)

Observability & Monitoring:
- session-logger.sh (JSONL activity logging)
- session-stats.sh (cost tracking & analysis)
- guide/observability.md (full documentation)

LLM-as-a-Judge Evaluation:
- output-evaluator.md agent (Haiku)
- /validate-changes command
- pre-commit-evaluator.sh (opt-in git hook)

Google Agent Whitepaper Integration:
- Context Triage Guide (Section 2.2.4)
- CLAUDE.md Injection Warning (Section 3.1.3)
- Agent Validation Checklist (Section 4.2.4)
- MCP Security: Tool Shadowing & Confused Deputy (Section 8.6)
- Session vs Memory patterns (Section 3.3.3)

Stats: 10 new files, 8 modified, 5 new guide sections

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-14 21:00:49 +01:00
Florian BRUNIAUX
19110eba22 feat(docs): add comprehensive data privacy documentation v3.2.0
- Create guide/data-privacy.md with retention policies (5y/30d/0)
- Add privacy notice to README.md
- Add section 2.6 "Data Flow & Privacy" to ultimate-guide.md
- Add Golden Rule #7 to cheatsheet.md (know what's sent)
- Add Phase 0.5 Privacy Awareness to onboarding-prompt.md
- Add privacy checks to audit-prompt.md
- Add PRIVACY CHECK section to audit-scan.sh (human + JSON)
- Add privacy reminder to check-claude.sh
- Create privacy-warning.sh SessionStart hook

Addresses user awareness of Anthropic data retention and opt-out options.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-14 20:42:19 +01:00
Florian BRUNIAUX
ebf1506196 fix(audit): ALL_DEPS unbound variable + extended integrations
- Initialize ALL_DEPS="" to prevent unbound variable error outside Node.js projects
- Added 35+ new integration detections: Kinde, LemonSqueezy, Vercel AI SDK,
  TalkJS, Knock, Stream, MapLibre, Mapbox, Google Maps, Bytescale, UploadThing,
  Cloudinary, Forest Admin, Refine, Vercel Analytics, Mixpanel, Hotjar, etc.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-13 08:42:49 +01:00
Florian BRUNIAUX
41bf57d9ee feat(audit): SSoT warning + incremental suggestions prompt
Script:
- needs_ssot_refactor flag: warns if CLAUDE.md >100 lines with 0 @refs
- Red warning in human output suggesting SSoT pattern

Prompt:
- Focus on incremental improvements, not generic advice
- Stricter health score (penalize missing SSoT)
- Domain-specific quick wins only
- Improve existing CLAUDE.md instead of full rewrite
- Don't suggest duplicates of existing agents/commands

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-12 15:08:06 +01:00
Florian BRUNIAUX
5b526f5e53 feat(audit): enhanced stack detection with integrations
- Detailed stack breakdown: runtime, framework, test, bundler, database
- Generic integration detection (25+ packages: Clerk, Stripe, OpenAI, Sentry...)
- jq fallback: grep-based JSON parsing when jq not installed
- Stack recap at top of human output + full stack object in JSON
- README prompt updated: Stack Recap first, CLAUDE.md template ~100 lines

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-12 14:52:55 +01:00
Florian BRUNIAUX
98044a0a47 fix(audit-scan): count files recursively in subfolders
- Commands in subfolders (tech/, product/, support/) now counted correctly
- Split count_files() into count_md_files() and count_script_files()
- Hooks counted as scripts (.sh/.js/.py/.ts), others as .md
- Excludes README.md from counts
- Bug: Was reporting 0 commands when 10 existed in subfolders
- Bump version to 2.9.6

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-12 14:20:53 +01:00
Florian BRUNIAUX
0833e1ca65 fix(audit): correct MCP detection to check ~/.claude.json
The audit script was looking for MCP config in ~/.claude/mcp.json which
doesn't exist. Claude Code actually stores MCP config in ~/.claude.json
under projects.<path>.mcpServers.

Changes:
- audit-scan.sh: Multi-source MCP detection (3 locations with priority)
- audit-scan.sh: Fixed count_pattern() bug causing "0\n0" output
- claude-setup-audit-prompt.md: Updated bash commands for MCP detection
- Version bump: 2.8 → 2.9

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
2026-01-12 08:37:01 +01:00
Florian BRUNIAUX
869f0df782 feat: optimize audit prompt with bash scanning (v2.7.0)
Add fast bash-based setup scanner for 80% speed improvement:
- New audit-scan.sh script with JSON/human output modes
- Rewrite audit prompt Phase 1-2 to use grep/find/wc
- Update README with dual-option audit approach
- Token reduction from ~5000 to ~500 (90% improvement)
- Execution time from ~30s to ~2s

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
2026-01-11 14:39:46 +01:00
Florian BRUNIAUX
76845f8226 chore: release v2.5.0 with content optimization and script externalization
- Remove ~1048 lines of non-Claude-Code-specific content (-10.9%)
- Externalize health check and reinstall scripts to examples/scripts/
- Clean up table of contents and fix broken references
- Update version numbers and statistics across all documentation

Removed sections:
- DeepSeek Integration (200 lines)
- Git Archaeology Pattern (250 lines)
- Emergency Hotfix Checklist (140 lines)
- Maturity Model & Success Metrics (95 lines)
- Generic Prompt Templates (105 lines)
- Task-specific checklists

New files:
- examples/scripts/check-claude.{sh,ps1}
- examples/scripts/clean-reinstall-claude.{sh,ps1}

Stats: 9593 → 8545 lines, focus on Claude Code-specific content

Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
2026-01-11 13:52:19 +01:00