Vodou Regression Test Suite

AI Agent Instructions

You ARE the test runner. When this skill loads:

1. Display the overview and STOPPING POINT 1 menu. STOP and wait for user selection.
2. After user selects scope, execute every test in the selected suite(s) sequentially.
3. For each test, run the exact command shown. Check exit code and output against the expected result.
4. Track results: maintain a running count of PASS, WARN, FAIL per suite.
5. After all selected suites complete, display the Results Report using the template at the bottom.
6. Display STOPPING POINT 2 with post-results options. STOP and wait.

Result classification:

• PASS: Command exits 0 AND output matches expected pattern
• WARN: Command works but output is degraded (e.g., empty table, optional file missing, slow but functional)
• FAIL: Command exits non-zero OR output contradicts expected pattern

Timing: Measure wall-clock time for every command. Include in performance summary.

Important: Run all commands from the Vodou installation root directory. All paths are relative to that root. Use cd to the Vodou root before starting if needed.

Compatibility: macOS ships with Bash 3.2. Do NOT use Bash 4+ features like declare -A (associative arrays), ${var,,} (lowercase), |& (pipe stderr), or &>> (append both). Use simple variables, arrays (declare -a is OK), and POSIX-compatible constructs. If you need structured tracking, use Python or individual counter variables.

Overview

This skill runs a comprehensive regression test against a fresh (or existing) Vodou installation. It systematically validates all 11 subsystems: binaries, database, MCP servers, intent routing, skills, memory, daemon/hooks, workspace, configuration, CLI commands, and performance.

Tests are split into two tiers:

• Tier A (No API Key): File checks, database queries, daemon ops -- works on fresh installs before configuring .env
• Tier B (API Key Required): MCP tool calls that connect to live servers

Total: ~81 tests across 11 suites.

STOPPING POINT 1: Select Test Scope

Choose what to test:

1. Run ALL Suites (Tier A + B) -- Full regression, requires API keys and running daemon
2. Run Tier A Only (No API Key) -- Offline tests: binaries, database, workspace, config, daemon
3. Suite 1: Binary & Architecture -- vodou-core, vodou-hook-bin, arch validation
4. Suite 2: Database & Migrations -- Tables, FTS5, schema
5. Suite 3: MCP Servers -- Server list, health-check, tool calls [Tier B]
6. Suite 4: Intent Routing -- Mapping count, critical intents [Partial Tier B]
7. Suite 5: Skills System -- Skill files, registry, loading via vodou-core [Partial Tier B]
8. Suite 6: Memory System -- Files, memory.db, daily logs, FTS5, embeddings, hybrid search, flush
9. Suite 7: Daemon & Hooks -- vodou-hook-bin ensure, daemon.sock, context
10. Suite 8: Workspace Bootstrap -- All workspace files, memory dir
11. Suite 9: Configuration -- .env, extractors.toml, config.json
12. Suite 10: Core CLI Commands -- list, tools, context, version [Partial Tier B]
13. Suite 11: Performance -- Timing benchmarks [Tier B]
14. Add/Change a test -- Add a new test, modify an existing one, or add a new suite to this skill

Reply with a number (1-14):

Suite 1: Binary & Architecture (Tier A)

Test 1.1: vodou-core exists and is executable

test -x ./vodou-core && echo "PASS" || echo "FAIL"

Expected: PASS

Test 1.2: vodou-hook-bin exists and is executable

test -x ./vodou-hook-bin && echo "PASS" || echo "FAIL"

Expected: PASS

Test 1.3: oi wrapper script exists and is executable

test -x ./do && echo "PASS" || echo "FAIL"

Expected: PASS

Test 1.4: vodou-core correct architecture

EXPECTED_ARCH=$(uname -m); file ./vodou-core | grep -qi "$(echo $EXPECTED_ARCH | sed 's/x86_64/x86.64/')" && echo "PASS: $EXPECTED_ARCH" || echo "FAIL: wrong arch for $EXPECTED_ARCH"

Expected: PASS matching host architecture (arm64 or x86_64)

Test 1.5: vodou-hook-bin correct architecture

EXPECTED_ARCH=$(uname -m); file ./vodou-hook-bin | grep -qi "$(echo $EXPECTED_ARCH | sed 's/x86_64/x86.64/')" && echo "PASS: $EXPECTED_ARCH" || echo "FAIL: wrong arch for $EXPECTED_ARCH"

Expected: PASS matching host architecture

Test 1.6: vodou-core version returns valid output

./vodou-core version

Expected: Exit code 0, output contains a version string

Suite 2: Database & Migrations (Tier A)

Test 2.1: vodou-core.db exists

test -f vodou-core.db && echo "PASS" || echo "FAIL"

Test 2.2: Core tables present

sqlite3 vodou-core.db ".tables" 2>/dev/null

Verify output contains ALL of these: mcp_servers, tools, intent_mappings, skills_registry, work_logs, scheduled_tasks, schema_version. PASS if all present, FAIL if any missing (list which ones). Note: memory tables live in memory.db, NOT vodou-core.db.

Test 2.3: MCP servers populated

sqlite3 vodou-core.db "SELECT COUNT(*) FROM mcp_servers"

Expected: >= 10. WARN if < 10, FAIL if 0.

Test 2.4: Tools populated

sqlite3 vodou-core.db "SELECT COUNT(*) FROM tools"

Expected: >= 30. WARN if < 30.

Test 2.5: memory.db exists

test -f memory.db && echo "PASS" || echo "WARN: memory.db not found (created by daemon on first run)"

Expected: PASS if daemon has run, WARN on fresh install (this is expected).

Suite 3: MCP Servers (Tier B)

Test 3.1: vodou-core list shows servers

./vodou-core list

Expected: Exit 0, output shows table with server names

Test 3.2: health-check

./vodou-core health-check

Expected: Exit 0. Count pass/fail. PASS if all pass, WARN if some fail, FAIL if all fail.

Test 3.3: Tool call - mcp-monitor get_cpu_info

./vodou-core call mcp-monitor get_cpu_info

Expected: Exit 0, returns CPU information

Test 3.4: Tool call - mcp-monitor get_memory_info

./vodou-core call mcp-monitor get_memory_info

Expected: Exit 0, returns memory information

Test 3.5: Node server dist/index.js files exist

for srv in Vodou-Enhanced-Thinking Vodou-script-executor Vodou-session-manager Vodou-channels Vodou-LLM-router; do
  test -f "MCP-servers/$srv/dist/index.js" && echo "PASS: $srv" || echo "FAIL: $srv missing dist/index.js"
done

Test 3.6: Python server entry points exist

test -f MCP-servers/uml-mcp/mcp_server.py && echo "PASS: UML" || echo "WARN: UML missing (optional)"

Test 3.7: Go binary exists (mcp-monitor)

test -x MCP-servers/mcp-monitor/bin/mcp-monitor && echo "PASS" || echo "FAIL: mcp-monitor binary missing"

Suite 4: Intent Routing (Tier A + Partial B)

Test 4.1: Intent mapping count (Tier A)

sqlite3 vodou-core.db "SELECT COUNT(*) FROM intent_mappings"

Expected: >= 400. WARN if 200-399, FAIL if < 200.

Test 4.2: Critical intents exist (Tier A)

sqlite3 vodou-core.db "SELECT keyword, server_name, tool_name FROM intent_mappings WHERE keyword='cpu'"
sqlite3 vodou-core.db "SELECT keyword, server_name, tool_name FROM intent_mappings WHERE keyword='disk'"
sqlite3 vodou-core.db "SELECT keyword, server_name, tool_name FROM intent_mappings WHERE keyword IN ('hello','oi hello') LIMIT 2"

Expected: cpu -> mcp-monitor::get_cpu_info, disk -> mcp-monitor::get_disk_info, hello -> vodou-core::vc_load_skill

Test 4.3: Skill intents have tool_parameters (Tier A)

sqlite3 vodou-core.db "SELECT keyword, tool_parameters FROM intent_mappings WHERE server_name='vodou-core' AND tool_name='vc_load_skill' AND tool_parameters IS NOT NULL LIMIT 3"

Expected: Results show skill_name in tool_parameters JSON

Test 4.4: brain command routes correctly (Tier B)

./vodou-core brain "cpu"

Expected: Exit 0, output shows CPU info routed through mcp-monitor

Suite 5: Skills System (Tier A + Partial B)

Test 5.1: Skills directory has SKILL.md files (Tier A)

find skills/ -name "SKILL.md" -not -path "*/.build/*" 2>/dev/null | wc -l

Expected: >= 40. WARN if 20-39, FAIL if < 20.

Test 5.2: skills_registry populated (Tier A)

sqlite3 vodou-core.db "SELECT COUNT(*) FROM skills_registry"

Expected: >= 40. WARN if < 40.

Test 5.3: hello skill loads via vodou-core (Tier B)

./vodou-core brain "hello"

Expected: Exit 0, output contains skill content for hello

Test 5.4: Skill intent mappings exist (Tier A)

sqlite3 vodou-core.db "SELECT COUNT(*) FROM intent_mappings WHERE server_name='vodou-core' AND tool_name='vc_load_skill'"

Expected: >= 20. WARN if < 20.

Test 5.5: Key skill intents have tool_parameters (Tier A)

sqlite3 vodou-core.db "SELECT keyword, tool_parameters FROM intent_mappings WHERE keyword IN ('hello','regression test','deep think') AND tool_parameters IS NOT NULL"

Expected: All three return rows with skill_name in tool_parameters JSON

Suite 6: Memory System (Tier A)

6A: Memory Files & Configuration

Test 6.1: MEMORY.md exists

test -f .vodou/workspace/MEMORY.md && echo "PASS" || echo "FAIL"

Test 6.2: memory.toml exists and has valid config

test -s .vodou/workspace/memory.toml && echo "PASS" || echo "FAIL"

Test 6.3: memory.toml has extraction provider set

grep -q '^\[extraction\]' .vodou/workspace/memory.toml && grep -q 'provider' .vodou/workspace/memory.toml && echo "PASS" || echo "WARN: extraction config missing"

Test 6.4: Daily memory log directory exists

test -d .vodou/workspace/memory && echo "PASS" || echo "FAIL"

Test 6.5: Daily memory logs present

COUNT=$(ls .vodou/workspace/memory/????-??-??.md 2>/dev/null | wc -l | tr -d ' '); echo "PASS: $COUNT daily logs"; [ "$COUNT" -eq 0 ] && echo "(WARN: no daily logs yet, expected on fresh install)"

Expected: PASS with count. WARN if 0 (fresh install).

Test 6.6: Today's daily log exists

TODAY=$(date +%Y-%m-%d); test -f ".vodou/workspace/memory/$TODAY.md" && echo "PASS: $TODAY.md exists" || echo "WARN: no log for today (normal if no prompts yet)"

6B: Memory Database (memory.db)

Test 6.7: memory.db exists

test -f memory.db && echo "PASS" || echo "WARN: memory.db not found (created by daemon on first run)"

Expected: PASS if daemon has run. WARN on fresh install.

Test 6.8: memory.db has expected tables

if test -f memory.db; then TABLES=$(sqlite3 memory.db ".tables" 2>/dev/null); for t in memory_chunks memory_embeddings memory_fts memory_cache; do echo "$TABLES" | grep -q "$t" && echo "PASS: $t" || echo "FAIL: $t missing from memory.db"; done; else echo "SKIP: memory.db not present"; fi

Test 6.9: memory.db chunk count

if test -f memory.db; then COUNT=$(sqlite3 memory.db "SELECT COUNT(*) FROM memory_chunks" 2>/dev/null); echo "PASS: $COUNT chunks in memory.db"; [ "$COUNT" -eq 0 ] && echo "(WARN: empty)"; else echo "SKIP: memory.db not present"; fi

Test 6.10: vodou-core.db has NO memory tables (cleaned up)

TABLES=$(sqlite3 vodou-core.db ".tables" 2>/dev/null); FOUND=0; for t in memory_chunks memory_embeddings memory_fts; do echo "$TABLES" | grep -q "$t" && { echo "FAIL: $t still in vodou-core.db (should be in memory.db only)"; FOUND=1; }; done; [ "$FOUND" -eq 0 ] && echo "PASS: no memory tables in vodou-core.db"

6C: FTS5 Full-Text Search (memory.db)

Test 6.11: FTS5 vtable functional (memory.db)

if test -f memory.db; then sqlite3 memory.db "SELECT COUNT(*) FROM memory_fts LIMIT 1" 2>/dev/null && echo "PASS" || echo "FAIL: FTS5 vtable broken"; else echo "SKIP: memory.db not present"; fi

Test 6.12: FTS5 search query executes

if test -f memory.db; then sqlite3 memory.db "SELECT COUNT(*) FROM memory_fts WHERE memory_fts MATCH 'test'" 2>/dev/null && echo "PASS" || echo "FAIL: FTS5 MATCH query failed"; else echo "SKIP: memory.db not present"; fi

Expected: Exit 0 (count can be 0 on fresh install)

Test 6.13: FTS5 index consistent with chunks

if test -f memory.db; then CHUNKS=$(sqlite3 memory.db "SELECT COUNT(*) FROM memory_chunks" 2>/dev/null); FTS=$(sqlite3 memory.db "SELECT COUNT(*) FROM memory_fts" 2>/dev/null); echo "Chunks: $CHUNKS, FTS rows: $FTS"; [ "$CHUNKS" = "$FTS" ] && echo "PASS: consistent" || echo "WARN: mismatch (chunks=$CHUNKS, fts=$FTS)"; else echo "SKIP: memory.db not present"; fi

6D: Vector Embeddings (memory.db)

Test 6.14: memory_embeddings table accessible

if test -f memory.db; then COUNT=$(sqlite3 memory.db "SELECT COUNT(*) FROM memory_embeddings" 2>/dev/null); echo "PASS: $COUNT embeddings in memory.db"; [ "$COUNT" -eq 0 ] && echo "(WARN: no embeddings yet)"; else echo "SKIP: memory.db not present"; fi

Test 6.15: Embeddings match chunk count

if test -f memory.db; then CHUNKS=$(sqlite3 memory.db "SELECT COUNT(*) FROM memory_chunks" 2>/dev/null); EMBEDS=$(sqlite3 memory.db "SELECT COUNT(*) FROM memory_embeddings" 2>/dev/null); echo "Chunks: $CHUNKS, Embeddings: $EMBEDS"; [ "$CHUNKS" = "$EMBEDS" ] && echo "PASS: 1:1 mapping" || echo "WARN: mismatch (some chunks may lack embeddings)"; else echo "SKIP: memory.db not present"; fi

Test 6.16: Embedding dimensions correct (384D AllMiniLML6V2)

if test -f memory.db && [ "$(sqlite3 memory.db "SELECT COUNT(*) FROM memory_embeddings" 2>/dev/null)" -gt 0 ]; then DIM=$(sqlite3 memory.db "SELECT LENGTH(embedding) FROM memory_embeddings LIMIT 1" 2>/dev/null); echo "Embedding blob size: $DIM bytes"; [ "$DIM" -gt 0 ] && echo "PASS: embeddings present" || echo "FAIL: empty embedding"; else echo "SKIP: no embeddings to check"; fi

6E: Memory CLI & Pipeline

Test 6.17: mem config returns valid configuration

OUTPUT=$(./vodou-core mem config 2>/dev/null); echo "$OUTPUT" | grep -q "memory_db" && echo "$OUTPUT" | grep -q "extraction_provider" && echo "PASS: config valid" || echo "FAIL: mem config missing expected fields"

Expected: Shows memory_db path, extraction_provider, workspace_root

Test 6.18: mem prompt accepts input and exits cleanly

echo "test memory recall" | ./vodou-core mem prompt 2>/dev/null; EXIT=$?; [ $EXIT -eq 0 ] && echo "PASS: mem prompt exit 0" || echo "FAIL: mem prompt exit $EXIT"

Expected: Exit 0. Memory recall is hook-driven — prompt is buffered and memories returned via daemon socket, not stdout.

Test 6.19: mem flush exits cleanly

./vodou-core mem flush 2>/dev/null; EXIT=$?; [ $EXIT -eq 0 ] && echo "PASS: mem flush exit 0" || echo "WARN: flush exit $EXIT (daemon may not be running, falls back to buffer)"

Expected: Exit 0 if daemon is running. Non-zero is WARN.

Test 6.20: vodou-hook-bin context returns memory data

OUTPUT=$(./vodou-hook-bin context 2>/dev/null); echo "$OUTPUT" | grep -qi "memory\|MEMORY\|workspace" && echo "PASS" || echo "WARN: context returned but no memory references"

Expected: Exit 0, output references memory or workspace files

Test 6.21: mem subcommands all present

HELP=$(./vodou-core mem --help 2>&1); PASS=0; TOTAL=0; for cmd in prompt flush setup promote promote-micro archive config test-extract; do TOTAL=$((TOTAL+1)); echo "$HELP" | grep -q "$cmd" && PASS=$((PASS+1)) || echo "FAIL: missing subcommand '$cmd'"; done; echo "PASS: $PASS/$TOTAL subcommands present"

Expected: 8/8 subcommands

Test 6.22: memory migration marker

test -f .vodou/.memory_migrated && echo "PASS: migration marker present" || echo "WARN: .memory_migrated not found (daemon may not have run migration yet)"

Test 6.23: Memory recall pipeline (end-to-end)

# Verify that memory chunks exist AND FTS can find them — proving the recall pipeline has data to work with
if test -f memory.db; then CHUNKS=$(sqlite3 memory.db "SELECT COUNT(*) FROM memory_chunks" 2>/dev/null); FTS_HIT=$(sqlite3 memory.db "SELECT COUNT(*) FROM memory_fts WHERE memory_fts MATCH 'memory OR test OR system'" 2>/dev/null); echo "Chunks: $CHUNKS, FTS hits: $FTS_HIT"; [ "$CHUNKS" -gt 0 ] && [ "$FTS_HIT" -gt 0 ] && echo "PASS: memory recall pipeline has searchable data" || echo "WARN: chunks=$CHUNKS, fts_hits=$FTS_HIT (may need more prompts)"; else echo "SKIP: memory.db not present"; fi

Expected: PASS with chunks > 0 and FTS hits > 0. This proves the full loop: prompts → extraction → storage → FTS indexing → searchable recall.

Suite 7: Daemon & Hooks (Tier A)

Test 7.1: vodou-hook-bin ensure starts daemon

./vodou-hook-bin ensure 2>/dev/null && echo "PASS" || echo "FAIL"

Expected: Exit 0

Test 7.2: daemon.sock exists

sleep 2 && test -S .vodou/daemon.sock && echo "PASS" || echo "FAIL: daemon.sock not found"

Test 7.3: daemon.pid valid

test -f .vodou/daemon.pid && kill -0 $(cat .vodou/daemon.pid) 2>/dev/null && echo "PASS: PID $(cat .vodou/daemon.pid)" || echo "FAIL: daemon not running"

Test 7.4: vodou-hook-bin context returns data

OUTPUT=$(./vodou-hook-bin context 2>/dev/null); [ -n "$OUTPUT" ] && echo "PASS" || echo "WARN: empty context (may be normal)"

Test 7.5: vodou-core daemon ensure

./vodou-core daemon ensure 2>&1 && echo "PASS" || echo "FAIL"

Expected: Exit 0

Suite 8: Workspace Bootstrap (Tier A)

Test 8.1-8.6: All workspace files exist

for f in MEMORY.md AGENTS.md USER.md IDENTITY.md SOUL.md TOOLS.md; do
  test -f ".vodou/workspace/$f" && echo "PASS: $f" || echo "FAIL: $f missing"
done

Test 8.7: memory directory exists

test -d .vodou/workspace/memory && echo "PASS" || echo "FAIL"

Test 8.8: memory.toml exists

test -f .vodou/workspace/memory.toml && echo "PASS" || echo "FAIL"

Suite 9: Configuration (Tier A)

Test 9.1: .env file exists

test -f .env && echo "PASS" || echo "FAIL"

Test 9.2: extractors.toml exists and not empty

test -s extractors.toml && echo "PASS" || echo "FAIL"

Test 9.3: mcp_servers registry exists in vodou-core.db

sqlite3 vodou-core.db "SELECT name FROM mcp_servers LIMIT 1" >/dev/null 2>&1 && echo "PASS" || echo "FAIL"

Test 9.4: mcp_servers has servers registered

sqlite3 vodou-core.db "SELECT 'PASS: ' || COUNT(*) || ' servers' FROM mcp_servers" 2>/dev/null || echo "FAIL"

Expected: >= 10 servers (this is the source of truth the LLM router reads; config.json is no longer generated)

Test 9.5: vodou-core.db schema version present

VER=$(sqlite3 vodou-core.db "SELECT MAX(version) FROM schema_version" 2>/dev/null); [ -n "$VER" ] && echo "PASS: schema version $VER" || echo "WARN: no schema_version table"

Expected: PASS with a version number. Pre-built DB should have schema_version seeded.

Suite 10: Core CLI Commands (Tier A + Partial B)

Test 10.1: list exits 0 (Tier B)

./vodou-core list 2>&1 && echo "PASS" || echo "FAIL"

Test 10.2: tools for mcp-monitor (Tier B)

./vodou-core tools mcp-monitor 2>&1 && echo "PASS" || echo "FAIL"

Test 10.3: context --base-only --json returns valid JSON (Tier A)

./vodou-core context --base-only --json 2>/dev/null | python3 -c "import sys,json; json.load(sys.stdin)" && echo "PASS" || echo "FAIL"

Test 10.4: version exits 0 (Tier A)

./vodou-core version 2>&1 && echo "PASS" || echo "FAIL"

Test 10.5: schedule list exits 0 (Tier A)

./vodou-core schedule list 2>&1 && echo "PASS" || echo "FAIL"

Test 10.6: list-tools-db exits 0 (Tier A)

./vodou-core list-tools-db 2>/dev/null && echo "PASS" || echo "FAIL"

Test 10.7: all-tools lists tools (Tier B)

./vodou-core all-tools 2>/dev/null && echo "PASS" || echo "FAIL"

Suite 11: Performance (Tier B)

For each test, measure wall-clock time and compare against the threshold.

Test 11.1: vodou-core version < 1 second

time ./vodou-core version 2>&1

Threshold: < 1s PASS, 1-3s WARN, > 3s FAIL

Test 11.2: vodou-hook-bin ensure < 3 seconds

time ./vodou-hook-bin ensure 2>&1

Threshold: < 3s PASS, 3-5s WARN, > 5s FAIL

Test 11.3: vodou-core list < 5 seconds

time ./vodou-core list 2>&1

Threshold: < 5s PASS, 5-10s WARN, > 10s FAIL

Test 11.4: Tool call < 10 seconds

time ./vodou-core call mcp-monitor get_cpu_info 2>&1

Threshold: < 10s PASS, 10-15s WARN, > 15s FAIL

Test 11.5: vodou-hook-bin context < 2 seconds

time ./vodou-hook-bin context 2>&1

Threshold: < 2s PASS, 2-5s WARN, > 5s FAIL

Results Report Template

After completing all selected suites, display results in this exact format:

====================================================
  Vodou REGRESSION TEST REPORT
====================================================
Date:         [YYYY-MM-DD HH:MM:SS TZ]
Vodou Version:   [from vodou-core version]
Architecture: [uname -m]
Test Scope:   [user's selection]
Vodou Root:      [pwd]

RESULTS BY SUITE
----------------------------------------------------
Suite 1: Binary & Architecture     [X/Y]  [PASS|WARN|FAIL]
Suite 2: Database & Migrations     [X/Y]  [PASS|WARN|FAIL]
Suite 3: MCP Servers               [X/Y]  [PASS|WARN|FAIL]
Suite 4: Intent Routing            [X/Y]  [PASS|WARN|FAIL]
Suite 5: Skills System             [X/Y]  [PASS|WARN|FAIL]
Suite 6: Memory System             [X/19]  [PASS|WARN|FAIL]
Suite 7: Daemon & Hooks            [X/Y]  [PASS|WARN|FAIL]
Suite 8: Workspace Bootstrap       [X/Y]  [PASS|WARN|FAIL]
Suite 9: Configuration             [X/Y]  [PASS|WARN|FAIL]
Suite 10: Core CLI Commands        [X/Y]  [PASS|WARN|FAIL]
Suite 11: Performance              [X/Y]  [PASS|WARN|FAIL]

MEMORY SYSTEM DETAIL
----------------------------------------------------
  Files & Config (6A)
    6.1  MEMORY.md exists              [PASS|FAIL]
    6.2  memory.toml valid             [PASS|FAIL]
    6.3  extraction provider set       [PASS|WARN]
    6.4  daily log directory           [PASS|FAIL]
    6.5  daily log count               [N logs] [PASS|WARN]
    6.6  today's daily log             [PASS|WARN]
  Memory Database (6B)
    6.7  memory.db exists              [PASS|WARN]
    6.8  memory.db tables              [PASS|FAIL|SKIP]
    6.9  memory.db chunk count         [N] [PASS|WARN|SKIP]
    6.10 no memory tables in bt4.db    [PASS|FAIL]
  FTS5 Full-Text Search (6C)
    6.11 FTS5 vtable functional        [PASS|FAIL|SKIP]
    6.12 FTS5 MATCH query              [PASS|FAIL|SKIP]
    6.13 FTS5/chunk consistency        [PASS|WARN|SKIP]
  Vector Embeddings (6D)
    6.14 embeddings accessible         [N] [PASS|WARN|SKIP]
    6.15 embeddings/chunk parity       [PASS|WARN|SKIP]
    6.16 embedding dimensions          [PASS|FAIL|SKIP]
  CLI & Pipeline (6E)
    6.17 mem config valid              [PASS|FAIL]
    6.18 mem prompt accepts input      [PASS|FAIL]
    6.19 mem flush exits cleanly       [PASS|WARN]
    6.20 vodou-hook-bin context (memory)  [PASS|WARN]
    6.21 mem subcommands present       [PASS|FAIL]
    6.22 migration marker              [PASS|WARN]
    6.23 memory recall pipeline (e2e)  [PASS|WARN|SKIP]

FAILURES (if any)
----------------------------------------------------
[Suite X] Test X.Y: [description]
  Command: [command that failed]
  Expected: [what was expected]
  Actual: [what happened]
  Fix: [how to fix]

WARNINGS (if any)
----------------------------------------------------
[Suite X] Test X.Y: [description]
  Note: [what was degraded and why it might be ok]

PERFORMANCE SUMMARY
----------------------------------------------------
vodou-core version:  [X.Xs]  [PASS|WARN|FAIL]
vodou-hook-bin ensure:    [X.Xs]  [PASS|WARN|FAIL]
vodou-core list:     [X.Xs]  [PASS|WARN|FAIL]
Tool call (cpu):       [X.Xs]  [PASS|WARN|FAIL]
vodou-hook-bin context:   [X.Xs]  [PASS|WARN|FAIL]

TOTALS
----------------------------------------------------
Passed:   [N]
Warnings: [N]
Failed:   [N]
Total:    [N]

STATUS: [OPERATIONAL | OPERATIONAL WITH WARNINGS | DEGRADED | BROKEN]
====================================================

Status determination:

• OPERATIONAL: 0 failures, 0 warnings
• OPERATIONAL WITH WARNINGS: 0 failures, >= 1 warning
• DEGRADED: 1-3 failures
• BROKEN: > 3 failures

STOPPING POINT 2: Post-Results Actions

What would you like to do next?

1. Re-run failed tests only -- Retry just the failures to see if they're intermittent
2. Run a different suite -- Go back to suite selection
3. Export results to file -- Save report to regression-test-results-[date].md in the Vodou root
4. Diagnose failures -- Get detailed troubleshooting for each failure
5. Auto-fix issues -- Attempt to automatically repair all failures and warnings
6. Run full suite again -- Complete re-run of all tests
7. Done -- Exit regression testing

Reply with a number (1-7):

Option 5 (Post-Results): Auto-Fix Issues

When the user selects option 5 from STOPPING POINT 2, the agent attempts to repair every FAIL and WARN from the test results.

Agent instructions for auto-fix:

1. Loop through all failures and warnings from the test results, in suite order.
2. For each issue, apply the appropriate fix from the table below.
3. Show a live log as you go: Fixing [Test X.Y]: [description]... [FIXED|SKIPPED|MANUAL]
4. After all fixes attempted, re-run only the previously failed/warned tests to verify.
5. Display a summary of what was fixed, what still needs attention, and return to STOPPING POINT 2.

Auto-fix actions by test:

Important:

• Always ask before running destructive fixes (deleting/recreating files that may have user data)
• MANUAL items: explain what the user needs to do and why it can't be automated
• If a fix fails, log it and move on -- don't retry in a loop
• After auto-fix, always re-run the affected tests to confirm the fix worked

Option 14: Add/Change a Test

When the user selects option 14, ask them:

What would you like to do?

1. Add a new test to an existing suite (1-11)
2. Modify an existing test (change command, expected result, or tier)
3. Remove a test
4. Add an entirely new suite (Suite 12+)

Reply with a number (1-4) and describe what you want:

Agent instructions for option 14:

• After the user describes the change, read this SKILL.md file, make the edit directly, and confirm what changed.
• For new tests: follow the existing format (**Test X.Y: description** + bash code block + Expected line). Number sequentially within the suite.
• For new suites: add a new ## Suite N: section, add it to STOPPING POINT 1 menu (increment option count), and add a line to the report template.
• Update the total test count in the Overview section.
• After editing, show the user the diff of what changed.

Troubleshooting Guide

Common Failures and Fixes

vodou-core wrong architecture:

• Re-download the correct binary for your platform (arm64 vs intel)
• Check: file ./vodou-core and compare with uname -m

Database tables missing:

• Run ./vodou-core list to trigger schema initialization
• Pre-built DB should have all tables; if missing, the build may be corrupt

MCP servers not responding:

• Run ./vodou-core health-check for per-server diagnostics
• Ensure daemon is running: ./vodou-hook-bin ensure
• Restart services: ./start-vodou-services.sh
• Check Node.js: node --version (must be >= 20)

FTS5 vtable broken (memory.db):

• Daemon auto-repairs on startup via ensure_fts_healthy()
• Force repair: restart daemon with ./vodou-core daemon ensure

memory.db missing or empty:

• Daemon creates memory.db on first run. Start daemon: ./vodou-hook-bin ensure
• All memory data lives exclusively in memory.db (not vodou-core.db)

FTS5/chunk count mismatch (memory.db):

• FTS index can drift from chunks after crashes or interrupted writes
• Fix: restart daemon (./vodou-core daemon ensure) -- ensure_fts_healthy() auto-repairs
• Nuclear option: sqlite3 memory.db "DELETE FROM memory_fts" then restart daemon

Embeddings missing or mismatched:

• Embeddings are generated by daemon during memory sync (AllMiniLML6V2, 384D)
• If chunks exist but embeddings don't: daemon hasn't processed them yet. Restart and wait.
• Check embedding model: grep -i embed .vodou/workspace/memory.toml

Legacy memory tables in vodou-core.db:

• If regression test 6.10 fails, legacy memory tables still exist in vodou-core.db
• Safe to drop: sqlite3 vodou-core.db "DROP TABLE IF EXISTS memory_fts; DROP TABLE IF EXISTS memory_chunks; DROP TABLE IF EXISTS memory_embeddings;"
• Note: migration 021 may recreate them on fresh installs; this is a known issue pending code cleanup

No daily memory logs:

• Daily logs are created by incremental extraction on each prompt (UserPromptSubmit hook)
• Verify hook is configured: check .claude/settings.json for vodou-hook-bin sock prompt
• Manual flush: ./vodou-hook-bin sock flush

vodou-hook-bin sock flush fails:

• Requires daemon to be running. Start with: ./vodou-hook-bin ensure
• If daemon is down, flush falls back to .prompt_buffer append (non-fatal)
• Check daemon status: test -S .vodou/daemon.sock && echo "running" || echo "down"

Workspace files missing:

• Run ./vodou-core bootstrap to recreate workspace files
• Check .vodou/workspace/ directory exists

mcp_servers registry empty (LLM router sees no servers):

• The router reads installed servers directly from vodou-core.db (mcp_servers table); config.json is no longer generated.
• Repopulate by reconnecting: ./vodou-core reconnect-all
• Verify: sqlite3 vodou-core.db "SELECT name FROM mcp_servers ORDER BY name"

Daemon won't start:

• Check for stale PID: cat .vodou/daemon.pid then ps aux | grep vodou-core
• Remove stale files: rm .vodou/daemon.pid .vodou/daemon.sock then retry
• Check logs: cat .vodou/daemon.log

better-sqlite3 ABI mismatch (Node servers fail):

• Run: ./scripts/rebuild-native-deps.sh
• Or per-server: cd MCP-servers/[server] && npm rebuild better-sqlite3