How long does the integration take?

Change one base URL in an OpenAI-compatible client, or install the Python or TypeScript SDK; the quickstart routes a first request in a few minutes.

Will this lock us into another vendor?

No. The router is Apache-2.0 open source, your provider keys stay yours, and the API is OpenAI-compatible, so leaving is a change of base URL.

What happens when a provider fails?

The router falls back to the next model your policy allows and records the fallback in the trace, so an outage costs a retry, not an incident.

validate-all - Prompt - OpenSmartRoute

Name: validate-all
Author: ItayAvioz

Prompt file imported from ItayAvioz/2026-World-Cup-Social-Predicting-App (.claude/commands/validate-all.md). Copyright stays with the author.

WorldCup 2026 — Master Validation Runner

ABSOLUTE RULE: READ-ONLY

All 6 agents are strictly read-only:

SQL: SELECT only — no INSERT, UPDATE, DELETE, DDL of any kind
Files: read only — no edits to migrations, memory, or SKILL.md
The ONLY write allowed is creating docs/VALIDATION_REPORT.md at the end
If an agent finds a problem → it reports it, never fixes it

Sources of Truth (leading — define correct behavior)

CLAUDE.md — app characterization, business rules, UI/UX rules
.claude/skills/db-feature/SKILL.md — live ERD, RLS spec, helper functions list
memory/db-phase.md — schema decisions, what's deployed, what's pending

Targets (agents verify everything else against the sources above)

supabase/migrations/*.sql — does migration code match the spec?
Live Supabase DB via SQL API — does the live DB match the migrations?
docs/*.md — are the docs in sync with the sources of truth?
test/*.html — do the test pages cover the full spec?

SQL Query Helper (for agents hitting the live DB — SELECT only)

PAT=$(cat "C:/Users/yonatanam/.claude/projects/C--Users-yonatanam-Desktop-World-Cup-APP/secrets/supabase_pat.txt")
curl -s -X POST "https://api.supabase.com/v1/projects/ftryuvfdihmhlzvbpfeu/database/query" \
  -H "Authorization: Bearer $PAT" -H "Content-Type: application/json" \
  --data-binary '{"query":"SQL_HERE"}'

Invocation Protocol

When /validate-all is invoked:

Step 1 — Read all 6 agent prompt files

.claude/skills/validate-all/agent-schema.md
.claude/skills/validate-all/agent-rls.md
.claude/skills/validate-all/agent-rules.md
.claude/skills/validate-all/agent-objects.md
.claude/skills/validate-all/agent-consistency.md
.claude/skills/validate-all/agent-pending.md

Step 2 — Spawn all 6 agents IN PARALLEL (background)

Use the Agent tool with run_in_background: true for all 6. Pass the full content of each agent file as the prompt. Use subagent_type: "Explore" for ALL agents.

WHY Explore: the Explore subagent type has Edit, Write, and NotebookEdit tools removed by design — agents literally cannot modify any file or schema. This is technical enforcement, not just a prompt instruction.

Step 3 — Wait for all 6 agents to complete

Do not proceed until all 6 background agents have returned results.

Step 4 — Collect and merge all 6 reports

Step 5 — Write merged report to `docs/VALIDATION_REPORT.md`

Step 6 — Print console summary (severity counts + top critical findings)

Repeatability Rule

Agents 1–4 (DB-facing) run every SQL check 3 times (R1, R2, R3):

All 3 identical → ✅ CONSISTENT — reliable
Any difference → ⚠️ FLAKY — report all 3 results, flag as timing/cron issue

Agents 5–6 (file-only) run each check once (no repeatability needed — files don't change mid-run).

Master Report Format (docs/VALIDATION_REPORT.md)

# WorldCup 2026 — Validation Report
Generated: [date]
Agents run: 6 | DB checks: 3× each | File checks: 1×

---

## Summary Table
| Agent          | Checks | ❌ Critical | ⚠️ Gap | ℹ️ Improvement | 📋 Decision |
|----------------|--------|------------|--------|----------------|-------------|
| Schema (A1)    |        |            |        |                |             |
| RLS (A2)       |        |            |        |                |             |
| Rules (A3)     |        |            |        |                |             |
| Objects (A4)   |        |            |        |                |             |
| Consistency(A5)|        |            |        |                |             |
| Pending (A6)   |        |            |        |                |             |
| **TOTAL**      |        |            |        |                |             |

---

## ❌ Critical Findings (fix before next phase)
[Each finding with full detail — ID, check, source, runs, fix recommendation]

## ⚠️ Gaps (important but not blocking)
[...]

## ℹ️ Improvements
[...]

## 📋 Decisions Needed
[...]

## ⚠️ Repeatability Issues (flaky checks)
[Any check that returned different results across 3 runs]

## Pending Items Priority Table
[From Agent 6 — full prioritized table]

---

## Gate: Ready for Edge Function Phase?
Based on findings:
- ❌ BLOCKED — [reason]
- ✅ CLEAR — all critical checks pass

Finding Format (each finding)

ID      | A1-003
Agent   | Schema
Check   | S-02: games — no status column
RUN 1   | [result]
RUN 2   | [result]
RUN 3   | [result]
MATCH   | ✅ CONSISTENT / ⚠️ FLAKY
STATUS  | ✅ PASS / ❌ FAIL / ⚠️ GAP / ℹ️ NOTE
SOURCE  | CLAUDE.md: "NO status column" / SKILL.md line 83
FINDING | [exact finding description]
FIX     | [recommended action — never performed by agents]

Agent Files Reference

File	Agent	Scope	DB calls?
agent-schema.md	A1	Tables, columns, constraints, FKs	YES — 3× each
agent-rls.md	A2	RLS enabled, policies, USING/WITH CHECK	YES — 3× each
agent-rules.md	A3	Business rules enforced in SQL	YES — 3× each
agent-objects.md	A4	RPCs, triggers, views, cron jobs	YES — 3× each
agent-consistency.md	A5	Cross-doc conflicts, outdated docs	NO — files only
agent-pending.md	A6	⚠️ items, TBDs, blockers, decisions	NO — files only

Install validate-all and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

# WorldCup 2026 — Master Validation Runner ## ABSOLUTE RULE: READ-ONLY All 6 agents are strictly read-only: - SQL: SELECT only — no INSERT, UPDATE, DELETE, DDL of any kind - Files: read only — no edits to migrations, memory, or SKILL.md - The ONLY write allowed is creating `docs/VALIDATION_REPORT.md` at the end - If an agent finds a problem → it reports it, never fixes it --- ## Sources of Truth (leading — define correct behavior) 1. `CLAUDE.md` — app characterization, business rules, UI/UX rules 2. `.claude/skills/db-feature/SKILL.md` — live ERD, RLS spec, helper functions list 3. `memory/db-phase.md` — schema decisions, what's deployed, what's pending ## Targets (agents verify everything else against the sources above) - `supabase/migrations/*.sql` — does migration code match the spec? - Live Supabase DB via SQL API — does the live DB match the migrations? - `docs/*.md` — are the docs in sync with the sources of truth? - `test/*.html` — do the test pages cover the full spec? --- ## SQL Query Helper (for agents hitting the live DB — SELECT only) ```bash PAT=$(cat "C:/Users/yonatanam/.claude/projects/C--Users-yonatanam-Desktop-World-Cup-APP/secrets/supabase_pat.txt") curl -s -X POST "https://api.supabase.com/v1/projects/ftryuvfdihmhlzvbpfeu/database/query" \ -H "Authorization: Bearer $PAT" -H "Content-Type: application/json" \ --data-binary '{"query":"SQL_HERE"}' ``` --- ## Invocation Protocol When `/validate-all` is invoked: ### Step 1 — Read all 6 agent prompt files ``` .claude/skills/validate-all/agent-schema.md .claude/skills/validate-all/agent-rls.md .claude/skills/validate-all/agent-rules.md .claude/skills/validate-all/agent-objects.md .claude/skills/validate-all/agent-consistency.md .claude/skills/validate-all/agent-pending.md ``` ### Step 2 — Spawn all 6 agents IN PARALLEL (background) Use the Agent tool with `run_in_background: true` for all 6. Pass the full content of each agent file as the prompt. Use `subagent_type: "Explore"` for ALL agents. WHY Explore: the Explore subagent type has Edit, Write, and NotebookEdit tools removed by design — agents literally cannot modify any file or schema. This is technical enforcement, not just a prompt instruction. ### Step 3 — Wait for all 6 agents to complete Do not proceed until all 6 background agents have returned results. ### Step 4 — Collect and merge all 6 reports ### Step 5 — Write merged report to `docs/VALIDATION_REPORT.md` ### Step 6 — Print console summary (severity counts + top critical findings) --- ## Repeatability Rule Agents 1–4 (DB-facing) run every SQL check **3 times** (R1, R2, R3): - All 3 identical → ✅ CONSISTENT — reliable - Any difference → ⚠️ FLAKY — report all 3 results, flag as timing/cron issue Agents 5–6 (file-only) run each check once (no repeatability needed — files don't change mid-run). --- ## Master Report Format (docs/VALIDATION_REPORT.md) ```markdown # WorldCup 2026 — Validation Report Generated: [date] Agents run: 6 | DB checks: 3× each | File checks: 1× --- ## Summary Table | Agent | Checks | ❌ Critical | ⚠️ Gap | ℹ️ Improvement | 📋 Decision | |----------------|--------|------------|--------|----------------|-------------| | Schema (A1) | | | | | | | RLS (A2) | | | | | | | Rules (A3) | | | | | | | Objects (A4) | | | | | | | Consistency(A5)| | | | | | | Pending (A6) | | | | | | | **TOTAL** | | | | | | --- ## ❌ Critical Findings (fix before next phase) [Each finding with full detail — ID, check, source, runs, fix recommendation] ## ⚠️ Gaps (important but not blocking) [...] ## ℹ️ Improvements [...] ## 📋 Decisions Needed [...] ## ⚠️ Repeatability Issues (flaky checks) [Any check that returned different results across 3 runs] ## Pending Items Priority Table [From Agent 6 — full prioritized table] --- ## Gate: Ready for Edge Function Phase? Based on findings: - ❌ BLOCKED — [reason] - ✅ CLEAR — all critical checks pass ``` --- ## Finding Format (each finding) ``` ID | A1-003 Agent | Schema Check | S-02: games — no status column RUN 1 | [result] RUN 2 | [result] RUN 3 | [result] MATCH | ✅ CONSISTENT / ⚠️ FLAKY STATUS | ✅ PASS / ❌ FAIL / ⚠️ GAP / ℹ️ NOTE SOURCE | CLAUDE.md: "NO status column" / SKILL.md line 83 FINDING | [exact finding description] FIX | [recommended action — never performed by agents] ``` --- ## Agent Files Reference | File | Agent | Scope | DB calls? | |---|---|---|---| | agent-schema.md | A1 | Tables, columns, constraints, FKs | YES — 3× each | | agent-rls.md | A2 | RLS enabled, policies, USING/WITH CHECK | YES — 3× each | | agent-rules.md | A3 | Business rules enforced in SQL | YES — 3× each | | agent-objects.md | A4 | RPCs, triggers, views, cron jobs | YES — 3× each | | agent-consistency.md | A5 | Cross-doc conflicts, outdated docs | NO — files only | | agent-pending.md | A6 | ⚠️ items, TBDs, blockers, decisions | NO — files only |

{ "prompt": "# WorldCup 2026 — Master Validation Runner\n\n## ABSOLUTE RULE: READ-ONLY\nAll 6 agents are strictly read-only:\n- SQL: SELECT only — no INSERT, UPDATE, DELETE, DDL of any kind\n- Files: read only — no edits to migrations, memory, or SKILL.md\n- The ONLY write allowed is creating `docs/VALIDATION_REPORT.md` at the end\n- If an agent finds a problem → it reports it, never fixes it\n\n---\n\n## Sources of Truth (leading — define correct behavior)\n1. `CLAUDE.md` — app characterization, business rules, UI/UX rules\n2. `.claude/skills/db-feature/SKILL.md` — live ERD, RLS spec, helper functions list\n3. `memory/db-phase.md` — schema decisions, what's deployed, what's pending\n\n## Targets (agents verify everything else against the sources above)\n- `supabase/migrations/*.sql` — does migration code match the spec?\n- Live Supabase DB via SQL API — does the live DB match the migrations?\n- `docs/*.md` — are the docs in sync with the sources of truth?\n- `test/*.html` — do the test pages cover the full spec?\n\n---\n\n## SQL Query Helper (for agents hitting the live DB — SELECT only)\n\n```bash\nPAT=$(cat \"C:/Users/yonatanam/.claude/projects/C--Users-yonatanam-Desktop-World-Cup-APP/secrets/supabase_pat.txt\")\ncurl -s -X POST \"https://api.supabase.com/v1/projects/ftryuvfdihmhlzvbpfeu/database/query\" \\\n -H \"Authorization: Bearer $PAT\" -H \"Content-Type: application/json\" \\\n --data-binary '{\"query\":\"SQL_HERE\"}'\n```\n\n---\n\n## Invocation Protocol\n\nWhen `/validate-all` is invoked:\n\n### Step 1 — Read all 6 agent prompt files\n```\n.claude/skills/validate-all/agent-schema.md\n.claude/skills/validate-all/agent-rls.md\n.claude/skills/validate-all/agent-rules.md\n.claude/skills/validate-all/agent-objects.md\n.claude/skills/validate-all/agent-consistency.md\n.claude/skills/validate-all/agent-pending.md\n```\n\n### Step 2 — Spawn all 6 agents IN PARALLEL (background)\nUse the Agent tool with `run_in_background: true` for all 6.\nPass the full content of each agent file as the prompt.\nUse `subagent_type: \"Explore\"` for ALL agents.\n\nWHY Explore: the Explore subagent type has Edit, Write, and NotebookEdit\ntools removed by design — agents literally cannot modify any file or schema.\nThis is technical enforcement, not just a prompt instruction.\n\n### Step 3 — Wait for all 6 agents to complete\nDo not proceed until all 6 background agents have returned results.\n\n### Step 4 — Collect and merge all 6 reports\n\n### Step 5 — Write merged report to `docs/VALIDATION_REPORT.md`\n\n### Step 6 — Print console summary (severity counts + top critical findings)\n\n---\n\n## Repeatability Rule\n\nAgents 1–4 (DB-facing) run every SQL check **3 times** (R1, R2, R3):\n- All 3 identical → ✅ CONSISTENT — reliable\n- Any difference → ⚠️ FLAKY — report all 3 results, flag as timing/cron issue\n\nAgents 5–6 (file-only) run each check once (no repeatability needed — files don't change mid-run).\n\n---\n\n## Master Report Format (docs/VALIDATION_REPORT.md)\n\n```markdown\n# WorldCup 2026 — Validation Report\nGenerated: [date]\nAgents run: 6 | DB checks: 3× each | File checks: 1×\n\n---\n\n## Summary Table\n| Agent | Checks | ❌ Critical | ⚠️ Gap | ℹ️ Improvement | 📋 Decision |\n|----------------|--------|------------|--------|----------------|-------------|\n| Schema (A1) | | | | | |\n| RLS (A2) | | | | | |\n| Rules (A3) | | | | | |\n| Objects (A4) | | | | | |\n| Consistency(A5)| | | | | |\n| Pending (A6) | | | | | |\n| **TOTAL** | | | | | |\n\n---\n\n## ❌ Critical Findings (fix before next phase)\n[Each finding with full detail — ID, check, source, runs, fix recommendation]\n\n## ⚠️ Gaps (important but not blocking)\n[...]\n\n## ℹ️ Improvements\n[...]\n\n## 📋 Decisions Needed\n[...]\n\n## ⚠️ Repeatability Issues (flaky checks)\n[Any check that returned different results across 3 runs]\n\n## Pending Items Priority Table\n[From Agent 6 — full prioritized table]\n\n---\n\n## Gate: Ready for Edge Function Phase?\nBased on findings:\n- ❌ BLOCKED — [reason]\n- ✅ CLEAR — all critical checks pass\n```\n\n---\n\n## Finding Format (each finding)\n\n```\nID | A1-003\nAgent | Schema\nCheck | S-02: games — no status column\nRUN 1 | [result]\nRUN 2 | [result]\nRUN 3 | [result]\nMATCH | ✅ CONSISTENT / ⚠️ FLAKY\nSTATUS | ✅ PASS / ❌ FAIL / ⚠️ GAP / ℹ️ NOTE\nSOURCE | CLAUDE.md: \"NO status column\" / SKILL.md line 83\nFINDING | [exact finding description]\nFIX | [recommended action — never performed by agents]\n```\n\n---\n\n## Agent Files Reference\n\n| File | Agent | Scope | DB calls? |\n|---|---|---|---|\n| agent-schema.md | A1 | Tables, columns, constraints, FKs | YES — 3× each |\n| agent-rls.md | A2 | RLS enabled, policies, USING/WITH CHECK | YES — 3× each |\n| agent-rules.md | A3 | Business rules enforced in SQL | YES — 3× each |\n| agent-objects.md | A4 | RPCs, triggers, views, cron jobs | YES — 3× each |\n| agent-consistency.md | A5 | Cross-doc conflicts, outdated docs | NO — files only |\n| agent-pending.md | A6 | ⚠️ items, TBDs, blockers, decisions | NO — files only |", "variables": [], "notes": "Read-only validation suite for WorldCup 2026. Checks DB schema, RLS, business rules, DB objects, doc consistency, and pending items against app characterization (CLAUDE.md), live ERD (SKILL.md), and memory files. Never modifies anything. Each DB check runs 3 times for repeatability. Outputs a full report to docs/VALIDATION_REPORT.md. Tools: Read, Bash, Grep, Glob, Agent, Write.", "metadata": { "source": { "provider": "github-claude-commands", "repository": "https://github.com/ItayAvioz/2026-World-Cup-Social-Predicting-App", "path": ".claude/commands/validate-all.md", "ref": "5dedd6e79c117c31663a68909590fa9e10517c28", "url": "https://github.com/ItayAvioz/2026-World-Cup-Social-Predicting-App/blob/5dedd6e79c117c31663a68909590fa9e10517c28/.claude/commands/validate-all.md", "key": "ItayAvioz/2026-World-Cup-Social-Predicting-App/.claude/commands/validate-all.md" } } }

validate-all

About

WorldCup 2026 — Master Validation Runner

ABSOLUTE RULE: READ-ONLY

Sources of Truth (leading — define correct behavior)

Targets (agents verify everything else against the sources above)

SQL Query Helper (for agents hitting the live DB — SELECT only)

Invocation Protocol

Step 1 — Read all 6 agent prompt files

Step 2 — Spawn all 6 agents IN PARALLEL (background)

Step 3 — Wait for all 6 agents to complete

Step 4 — Collect and merge all 6 reports

Step 5 — Write merged report to `docs/VALIDATION_REPORT.md`

Step 6 — Print console summary (severity counts + top critical findings)

Repeatability Rule

Master Report Format (docs/VALIDATION_REPORT.md)

Finding Format (each finding)

Agent Files Reference

Install validate-all and it becomes one more candidate the router can pick - when it fits.

What it costs to adopt - in engineering time.

One base URL

No vendor lock-in

Failover built in

Use it

Manifest

Reviews

About

WorldCup 2026 — Master Validation Runner

ABSOLUTE RULE: READ-ONLY

Sources of Truth (leading — define correct behavior)

Targets (agents verify everything else against the sources above)

SQL Query Helper (for agents hitting the live DB — SELECT only)

Invocation Protocol

Step 1 — Read all 6 agent prompt files

Step 2 — Spawn all 6 agents IN PARALLEL (background)

Step 3 — Wait for all 6 agents to complete

Step 4 — Collect and merge all 6 reports

Step 5 — Write merged report to docs/VALIDATION_REPORT.md

Step 6 — Print console summary (severity counts + top critical findings)

Repeatability Rule

Master Report Format (docs/VALIDATION_REPORT.md)

Finding Format (each finding)

Agent Files Reference

Install validate-all and it becomes one more candidate the router can pick - when it fits.

What it costs to adopt - in engineering time.

One base URL

No vendor lock-in

Failover built in

Use it

Manifest

Reviews

Step 5 — Write merged report to `docs/VALIDATION_REPORT.md`