tri

name	tri
description	TRI status dashboard — agent thought mode by default, /tri full for visual dashboard

For system state collection, follow .claude/skills/_shared/data_collection.md. For output formatting conventions, follow .claude/skills/_shared/output_format.md. For cached system state, follow .claude/skills/_shared/system_snapshot.md.

Mode Detection

Check $ARGUMENTS for mode:

• If $ARGUMENTS contains "full" → MODE=FULL (strip "full" from arguments, pass rest through)
• If $ARGUMENTS contains "short" → MODE=COMPACT (explicit short/conversational mode)
• If $ARGUMENTS contains "coverage" → MODE=COVERAGE (CLI/MCP coverage + duplicates + board audit)
• Otherwise → MODE=COMPACT

"audit" and "lang:xx" work in BOTH modes.

Run a diagnostic of the TRI system. In COMPACT mode, output ~15 lines. In FULL mode, output the complete diagnostic report.

If $ARGUMENTS contains "lang:ru" or "lang:en", update .claude/skills/tri/lang.md to that language and use it. If $ARGUMENTS is "ru" or "en" alone, treat as language switch. In FULL mode, if remaining $ARGUMENTS is provided, focus the diagnostic on that area.

Mood Detection

If $ARGUMENTS contains "humor", "prof", "toxic", or "zen" → set MOOD to that value and update .claude/skills/tri/mode.md with mood: {value}. If no mood keyword in $ARGUMENTS → read current mood from .claude/skills/tri/mode.md (default: humor).

🔄 Audit Mode

If $ARGUMENTS contains "audit", run a FRESH regeneration audit BEFORE the diagnostic:

python3 << 'AUDIT_EOF'
import subprocess, os, glob, time
specs = sorted(glob.glob('specs/tri/*.tri'))
results = []
for s in specs:
    name = os.path.splitext(os.path.basename(s))[0]
    try: subprocess.run(['zig-out/bin/vibee', 'gen', s], capture_output=True, timeout=30)
    except: pass
    zig = f'generated/{name}.zig'
    if not os.path.exists(zig):
        results.append((name, 'no output'))
        continue
    try:
        r = subprocess.run(['zig', 'ast-check', zig], capture_output=True, timeout=30)
        results.append((name, 'pass' if r.returncode == 0 else 'ast-check failed'))
    except: results.append((name, 'timeout'))
passed = sum(1 for _,s in results if s == 'pass')
total = len(results)
pct = passed*100//total if total else 0
emoji = '💎' if pct>=80 else ('🟡' if pct>=50 else '💀')
with open('specs/REGENERATION_REPORT.md','w') as f:
    f.write(f'# Regeneration Audit Report\n\n**Date:** {int(time.time())}\n**Sample:** {total} specs\n**Tool:** vibee gen + zig ast-check\n\n## Results\n\n| # | Spec | Status |\n|---|------|--------|\n')
    for i,(name,status) in enumerate(results,1):
        icon = '✅' if status=='pass' else f'❌ {status}'
        f.write(f'| {i} | {name} | {icon} |\n')
    f.write(f'\n## Summary\n\n- **Compiled:** {passed}/{total} = **{pct}%** {emoji}\n- **Failed:** {total-passed}\n')
print(f'AUDIT_COMPLETE: {passed}/{total} = {pct}%')
AUDIT_EOF

Then proceed with normal diagnostic report using fresh data.

📊 Coverage Mode (when MODE=COVERAGE)

If MODE=COVERAGE, render ONLY this section, then STOP. Do not continue to other modes.

Step 1: Extract CLI commands

cd /Users/playra/trinity-w1 && sed -n '/^pub fn parseCommand/,/^}/p' src/tri/tri_utils.zig | grep -oE '"[a-z][a-z0-9_-]*"' | tr -d '"' | sort -u

Step 2: Extract MCP tools

cd /Users/playra/trinity-w1 && grep -o '"name":"[^"]*"' tools/mcp/trinity_mcp/server.zig | sed 's/"name":"//;s/"//' | sort -u

Step 3: Duplicate detection

cd /Users/playra/trinity-w1 && \
echo "=== DELEGATE_CLI ===" && grep -c 'executeTriSimple' tools/mcp/trinity_mcp/server.zig && \
echo "=== DELEGATE_MODULE ===" && grep -cE 'swarm\.\w+\(|cloud_orch\.\w+|chain_engine|needle\.' tools/mcp/trinity_mcp/server.zig && \
echo "=== TOTAL_TOOLS ===" && grep -o '"name":"[^"]*"' tools/mcp/trinity_mcp/server.zig | wc -l

Step 4: GitHub board sync status

cd /Users/playra/trinity-w1 && \
gh issue list -R gHashTag/trinity --state open --limit 20 --json number,title,labels --jq '.[] | select(.labels[].name == "status:in-progress") | "\(.number) \(.title)"' 2>&1 && \
echo "=== BOARD ===" && gh project item-list 6 --owner gHashTag --format json --limit 50 2>&1

Step 5: Render report

Output a 3-paragraph report:

P1: Coverage Score — {N} CLI commands, {N} MCP tools, {coverage}% coverage.

P2: Gaps & Duplicates — Top 5 CLI commands without MCP exposure. Top 5 MCP tools without CLI counterpart. Dedup score.

P3: Board Sync — Open issues count, in-progress count, board items count, out-of-sync count.

🌐 Language System

Before rendering the report, read .claude/skills/tri/lang.md to determine the output language. The file contains lang: ru or lang: en. Default: en.

All section headers, labels, problem descriptions, oracle commentary, and φ-liners MUST be rendered in the chosen language. Technical terms stay in English.

For the complete translation table, follow .claude/skills/_shared/language.md.

---

Routing to Shared Modules

IF MODE=COMPACT:

Follow .claude/skills/_shared/compact_narration.md. STOP after completion.

IF MODE=FULL:

1. Follow .claude/skills/_shared/full_diagnostics.md — collect all data
2. Follow .claude/skills/_shared/report_template.md — render tables and faculty
3. Follow .claude/skills/_shared/oracle_verdict.md — oracle, Loop-0, action plan, benchmarks