Terraphim AI documentation

Design & Implementation Plan: Terraphim Knowledge Graph Workflows

1. Summary of Target Behavior

After implementation, Terraphim will provide a complete local-first knowledge graph validation pipeline for AI coding workflows:

Pre-LLM Validation

  • Before sending queries to LLMs, validate that input terms are semantically connected
  • Suggest fuzzy alternatives when exact terms aren't found
  • Apply role-specific knowledge graphs for domain validation

Post-LLM Validation

  • Verify LLM outputs against domain checklists stored in knowledge graph
  • Extract relevant concepts and validate terminology compliance
  • Flag outputs that use non-standard terms

Smart Commit Integration

  • Extract concepts from changed files for commit message enrichment
  • Validate commit messages against project knowledge graph

Unified CLI & Hook Interface

  • All features accessible via terraphim-agent subcommands
  • Role selection via --role flag across all commands
  • Single hook entry point for Claude Code integration

2. Key Invariants and Acceptance Criteria

Invariants (Must Always Hold)

| ID | Invariant | Enforcement | |----|-----------|-------------| | I1 | Hooks complete in <200ms for typical inputs | Timeout + early exit | | I2 | All validation is local-first (no network required) | Use only local KG files | | I3 | Existing hooks continue to work unchanged | Backward-compatible CLI | | I4 | Role graphs are loaded lazily | Only load when role is accessed | | I5 | Connectivity check limits to ≤10 matched terms | Hard limit with warning |

Acceptance Criteria

| ID | Criterion | Test Type | |----|-----------|-----------| | AC1 | terraphim-agent validate --connectivity "text" returns true/false with matched terms | Unit | | AC2 | terraphim-agent suggest --fuzzy "typo" returns top 5 suggestions with similarity scores | Unit | | AC3 | terraphim-agent replace --role "X" "text" uses role X's thesaurus | Integration | | AC4 | terraphim-agent extract --paragraphs "text" returns matched term + paragraph pairs | Unit | | AC5 | terraphim-agent validate --checklist "output" validates against domain checklist | Integration | | AC6 | Pre-LLM hook enriches context with KG concepts before LLM call | E2E | | AC7 | Post-LLM hook validates output and adds warnings if terms not in KG | E2E | | AC8 | MCP is_all_terms_connected_by_path calls real RoleGraph implementation | Integration | | AC9 | Smart commit extracts concepts from git diff and suggests commit message elements | E2E |

3. High-Level Design and Boundaries

Architecture Overview

┌─────────────────────────────────────────────────────────────────────────────┐
│                           Skills Layer (New)                                 │
│  ┌───────────────────┐  ┌───────────────────┐  ┌───────────────────┐       │
│  │ pre-llm-validate  │  │ post-llm-check    │  │ smart-commit      │       │
│  │ (skill file)      │  │ (skill file)      │  │ (skill file)      │       │
│  └─────────┬─────────┘  └─────────┬─────────┘  └─────────┬─────────┘       │
└────────────┼────────────────────────┼────────────────────────┼──────────────┘
             │                        │                        │
             ▼                        ▼                        ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│                           Hooks Layer (Updated)                              │
│  ┌───────────────────┐  ┌───────────────────┐  ┌───────────────────┐       │
│  │ pre-tool-use.sh   │  │ post-tool-use.sh  │  │ prepare-commit.sh │       │
│  │ (calls agent)     │  │ (calls agent)     │  │ (calls agent)     │       │
│  └─────────┬─────────┘  └─────────┬─────────┘  └─────────┬─────────┘       │
└────────────┼────────────────────────┼────────────────────────┼──────────────┘
             │                        │                        │
             ▼                        ▼                        ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│                      terraphim-agent CLI (Extended)                          │
│                                                                              │
│  Existing Commands:        New Commands:                                     │
│  ┌──────────┐             ┌──────────────┐  ┌──────────────┐               │
│  │ replace  │ ──extend──▶ │ validate     │  │ suggest      │               │
│  │ extract  │             │ --connectivity│  │ --fuzzy      │               │
│  │ search   │             │ --checklist  │  │ --threshold  │               │
│  └──────────┘             │ --role       │  │ --role       │               │
│                           └──────────────┘  └──────────────┘               │
│                                                                              │
│  New Subcommand:          Extended:                                          │
│  ┌──────────────┐        ┌──────────────┐                                   │
│  │ hook         │        │ replace      │                                   │
│  │ --type X     │        │ --role X     │                                   │
│  │ --input JSON │        │ --suggest    │                                   │
│  └──────────────┘        └──────────────┘                                   │
└─────────────────────────────────────────────────────────────────────────────┘
             │
             ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│                      terraphim_tui/src/ (CLI Implementation)                 │
│                                                                              │
│  ┌─────────────────────────────────────────────────────────────────────┐   │
│  │                     commands/ (New Module)                           │   │
│  │  ┌──────────────┐  ┌──────────────┐  ┌──────────────┐              │   │
│  │  │ validate.rs  │  │ suggest.rs   │  │ hook.rs      │              │   │
│  │  └──────────────┘  └──────────────┘  └──────────────┘              │   │
│  └─────────────────────────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────────────────────────┘
             │
             ▼
┌─────────────────────────────────────────────────────────────────────────────┐
│                         Core Crates (Minimal Changes)                        │
│                                                                              │
│  terraphim_mcp_server:     terraphim_rolegraph:    terraphim_automata:      │
│  ┌──────────────────┐     ┌──────────────────┐    ┌──────────────────┐     │
│  │ Fix connectivity │     │ No changes       │    │ No changes       │     │
│  │ placeholder      │     │ (already works)  │    │ (already works)  │     │
│  └──────────────────┘     └──────────────────┘    └──────────────────┘     │
└─────────────────────────────────────────────────────────────────────────────┘

Component Boundaries

| Component | Responsibility | Changes | |-----------|---------------|---------| | Skills (~/.claude/plugins/) | Workflow orchestration, user-facing patterns | New skill files | | Hooks (.claude/hooks/) | Claude Code integration, JSON I/O | Extend existing, add new | | terraphim-agent CLI | Feature exposure, argument parsing | New subcommands | | terraphim_tui | CLI implementation | New command modules | | terraphim_mcp_server | MCP tool exposure | Fix connectivity placeholder | | terraphim_rolegraph | Core graph operations | No changes needed | | terraphim_automata | Core text matching | No changes needed |

4. File/Module-Level Change Plan

4.1 MCP Connectivity Fix

| File | Action | Before | After | Dependencies | |------|--------|--------|-------|--------------| | crates/terraphim_mcp_server/src/lib.rs | Modify | Placeholder returns matched terms only | Calls RoleGraph::is_all_terms_connected_by_path | terraphim_rolegraph | | crates/terraphim_mcp_server/tests/test_advanced_automata_functions.rs | Modify | Tests expect placeholder behavior | Tests verify actual connectivity result | None |

4.2 CLI Commands

| File | Action | Before | After | Dependencies | |------|--------|--------|-------|--------------| | crates/terraphim_tui/src/commands/mod.rs | Create | - | Module exports for new commands | None | | crates/terraphim_tui/src/commands/validate.rs | Create | - | validate subcommand with --connectivity, --checklist, --role | terraphim_rolegraph | | crates/terraphim_tui/src/commands/suggest.rs | Create | - | suggest subcommand with --fuzzy, --threshold, --role | terraphim_automata | | crates/terraphim_tui/src/commands/hook.rs | Create | - | hook subcommand with --type, --input for unified hook handling | All core crates | | crates/terraphim_tui/src/main.rs | Modify | Current subcommands | Add new subcommand routing | commands module | | crates/terraphim_tui/src/replace.rs | Modify | No --role flag | Add --role and --suggest flags | terraphim_config |

4.3 Skills

| File | Action | Before | After | Dependencies | |------|--------|--------|-------|--------------| | skills/pre-llm-validate/skill.md | Create | - | Pre-LLM validation workflow skill | terraphim-agent CLI | | skills/post-llm-check/skill.md | Create | - | Post-LLM checklist validation skill | terraphim-agent CLI | | skills/smart-commit/skill.md | Create | - | Commit message enrichment skill | terraphim-agent CLI |

4.4 Hooks

| File | Action | Before | After | Dependencies | |------|--------|--------|-------|--------------| | .claude/hooks/pre-llm-validate.sh | Create | - | Calls terraphim-agent validate --connectivity | terraphim-agent | | .claude/hooks/post-llm-check.sh | Create | - | Calls terraphim-agent validate --checklist | terraphim-agent | | .claude/hooks/prepare-commit-msg | Modify | Basic replacement | Add concept extraction via terraphim-agent extract | terraphim-agent | | .claude/hooks/npm_to_bun_guard.sh | Modify | Hardcoded role | Use --role from env or config | terraphim-agent |

4.5 Knowledge Graph Extensions

| File | Action | Before | After | Dependencies | |------|--------|--------|-------|--------------| | docs/src/kg/checklists/ | Create | - | Directory for domain checklists | None | | docs/src/kg/checklists/code_review.md | Create | - | Code review checklist as KG | None | | docs/src/kg/checklists/security.md | Create | - | Security validation checklist as KG | None |

5. Step-by-Step Implementation Sequence

Phase A: Foundation (Fix MCP, Add CLI Infrastructure)

| Step | Purpose | Deployable? | Notes | |------|---------|-------------|-------| | A1 | Fix MCP connectivity placeholder to call real RoleGraph | Yes | Critical blocker | | A2 | Update MCP tests to verify actual connectivity | Yes | Validates A1 | | A3 | Create commands/ module structure in terraphim_tui | Yes | Infrastructure | | A4 | Add --role flag to existing replace command | Yes | Backward compatible |

Phase B: New CLI Commands

| Step | Purpose | Deployable? | Notes | |------|---------|-------------|-------| | B1 | Implement validate --connectivity command | Yes | Core feature | | B2 | Implement suggest --fuzzy command | Yes | Core feature | | B3 | Implement validate --checklist command | Yes | Requires B1 | | B4 | Implement hook unified handler command | Yes | Simplifies hooks | | B5 | Add unit tests for all new commands | Yes | Quality gate |

Phase C: Skills & Hooks

| Step | Purpose | Deployable? | Notes | |------|---------|-------------|-------| | C1 | Create pre-llm-validate skill | Yes | Uses B1 | | C2 | Create pre-llm-validate.sh hook | Yes | Integrates C1 | | C3 | Create post-llm-check skill | Yes | Uses B3 | | C4 | Create post-llm-check.sh hook | Yes | Integrates C3 | | C5 | Update prepare-commit-msg with concept extraction | Yes | Uses existing extract | | C6 | Create smart-commit skill | Yes | Orchestrates C5 |

Phase D: Knowledge Graph Extensions

| Step | Purpose | Deployable? | Notes | |------|---------|-------------|-------| | D1 | Create checklists/ directory structure | Yes | Infrastructure | | D2 | Create code_review checklist KG | Yes | Example checklist | | D3 | Create security checklist KG | Yes | Example checklist | | D4 | Document checklist format in docs | Yes | User guidance |

Phase E: Integration & Documentation

| Step | Purpose | Deployable? | Notes | |------|---------|-------------|-------| | E1 | Update CLAUDE.md with new skills/hooks | Yes | Discovery | | E2 | Create integration tests for full workflows | Yes | E2E validation | | E3 | Update install-terraphim-hooks.sh | Yes | Easy onboarding | | E4 | Performance benchmark hooks | Yes | Validate I1 invariant |

6. Testing & Verification Strategy

Unit Tests

| Acceptance Criterion | Test Location | Description | |---------------------|---------------|-------------| | AC1 (validate --connectivity) | terraphim_tui/tests/validate_test.rs | Test connected/disconnected text cases | | AC2 (suggest --fuzzy) | terraphim_tui/tests/suggest_test.rs | Test typo suggestions, threshold variations | | AC3 (replace --role) | terraphim_tui/tests/replace_test.rs | Test role-specific thesaurus selection | | AC4 (extract --paragraphs) | Existing tests | Already covered in terraphim_automata |

Integration Tests

| Acceptance Criterion | Test Location | Description | |---------------------|---------------|-------------| | AC5 (validate --checklist) | terraphim_tui/tests/checklist_test.rs | Test against sample checklists | | AC8 (MCP connectivity) | terraphim_mcp_server/tests/ | Update existing tests |

E2E Tests

| Acceptance Criterion | Test Location | Description | |---------------------|---------------|-------------| | AC6 (pre-LLM hook) | tests/e2e/pre_llm_hook_test.sh | Full hook invocation with sample input | | AC7 (post-LLM hook) | tests/e2e/post_llm_hook_test.sh | Full hook invocation with LLM output | | AC9 (smart commit) | tests/e2e/smart_commit_test.sh | Git diff to enriched commit message |

Performance Tests

| Invariant | Test | Threshold | |-----------|------|-----------| | I1 (hook latency) | benches/hook_latency.rs | <200ms p99 | | I5 (term limit) | tests/validate_term_limit_test.rs | Warning at >10 terms |

7. Risk & Complexity Review

| Risk (from Phase 1) | Mitigation in Design | Residual Risk | |---------------------|---------------------|---------------| | Connectivity check too slow | Hard limit of 10 terms (I5), timeout in hook | Low - bounded complexity | | MCP fix breaks existing tests | Step A2 updates tests alongside fix | Low - tested together | | Role loading increases startup | Lazy loading (I4) in CLI commands | Low - on-demand only | | Paragraph extraction misses code | Out of scope for v1, document limitation | Medium - future enhancement | | Pre-LLM validation too strict | Skills use advisory mode (warnings, not blocking) | Low - user control | | Hook complexity confuses users | Unified hook command, clear docs | Low - simplified interface |

Complexity Assessment

| Component | Complexity | Justification | |-----------|------------|---------------| | MCP fix | Low | Single function replacement | | CLI commands | Medium | New module structure, argument parsing | | Skills | Low | Markdown files with workflow docs | | Hooks | Low | Shell scripts calling CLI | | Checklists | Low | Markdown KG files |

Total estimated complexity: Medium (mostly additive, minimal changes to core crates)

8. Open Questions / Decisions for Human Review

Design Decisions Needed

  1. Pre-LLM mode: Should validation be advisory (add warnings) or blocking (reject)?

    • Recommendation: Advisory by default, blocking opt-in via --strict flag

  2. Role detection: How should hooks determine which role to use?

    • Recommendation: Priority order: --role flag > TERRAPHIM_ROLE env > project config > default

  3. Checklist format: Should checklists use existing KG synonyms format or new checklist:: directive?

    • Recommendation: New checklist:: directive for explicit semantics

  4. Hook timeout: What's the acceptable timeout for hook operations?

    • Recommendation: 200ms default, configurable via TERRAPHIM_HOOK_TIMEOUT

Scope Confirmation

  1. Smart commit scope: Should commit enrichment be automatic or skill-invoked?

    • Recommendation: Skill-invoked initially, automatic as optional future enhancement

  2. Existing skill updates: Update terraphim-hooks skill or create separate skills?

    • Recommendation: Create separate focused skills, update terraphim-hooks to reference them

Appendix: Proposed CLI Interface

# Validate semantic connectivity
terraphim-agent validate --connectivity "system operator trained for life cycle"
# Output: { "connected": true, "terms": [...], "path_exists": true }

terraphim-agent validate --connectivity --role "Security" "authentication protocol"
# Output: Uses Security role's knowledge graph

# Validate against checklist
terraphim-agent validate --checklist code_review "implemented feature with tests"
# Output: { "passed": ["has_tests"], "missing": ["security_check", "docs"] }

# Fuzzy suggestions
terraphim-agent suggest --fuzzy "terraphm" --threshold 0.7
# Output: [{ "term": "terraphim", "similarity": 0.92 }, ...]

# Role-aware replacement
terraphim-agent replace --role "DevOps" "run npm install"
# Output: "run bun install" (using DevOps role's thesaurus)

# Unified hook handler
terraphim-agent hook --type pre-tool-use --input '{"command": "npm test"}'
# Output: Processed JSON for Claude Code

# Extract concepts for commit
terraphim-agent extract --paragraphs --from-diff HEAD~1
# Output: Matched concepts from changed files

Do you approve this plan as-is, or would you like to adjust any part?

Sentrux Quality Gates Implementation Summary: Knowledge Graph Validation Workflows