Pipes SDK

One skill for the full Pipes SDK lifecycle: scaffold new indexers, diagnose runtime errors, optimize sync performance, and validate data quality.

When to Use This Skill

Activate when the user wants to:

• Create a new indexer — scaffold EVM or Solana projects with the CLI; set up Tron, Bitcoin, or Hyperliquid indexers manually with the SDK
• Fix runtime errors — compilation failures, DB issues, Portal timeouts, decoding errors
• Optimize sync performance — slow indexing, high memory, large ranges
• Validate data quality — NULL checks, gaps, malformed addresses, duplicate events
• Deploy an indexer — local Docker or ClickHouse Cloud
• Write to analytics sinks: ClickHouse, PostgreSQL (Drizzle), BigQuery, Parquet files, Google Pub/Sub
• Turn a Portal handoff into durable infrastructure — when a Portal MCP answer or curl export is not enough

Common trigger phrases: "create a new indexer", "my indexer crashed", "error", "not working", "slow", "optimize", "deploy to ClickHouse Cloud", "track X events on Ethereum/Solana/Tron/Bitcoin/Hyperliquid".

Portal Handoff Workflow

Use Pipes when a Portal MCP or raw Stream API workflow crosses from exploration into a maintained data product.

Good Pipes triggers:

• long historical backfills
• recurring syncs or scheduled refresh
• protocol-specific joins
• normalized tables for an app/backend
• custom metrics, alerts, or dashboards
• transforms that should be versioned and tested
• data too large for MCP chat responses or one-off curl files

If the input includes a Portal pipes_handoff recipe, preserve its network, filters, time window, outputs, and validation hints. Treat the Portal MCP/curl result as the baseline, scaffold the indexer, and validate the first output against the closest Portal query over the same bounded window.

Before scaffolding, say why Pipes is the right surface. Portal MCP is best for bounded answers and investigation pivots; Portal Stream API/curl is best for reproducible one-off extraction; Pipes is best for durable indexing, transforms, and storage.

Critical Environment Constraints

Node.js: v22 LTS. @subsquid/pipes declares engines.node >= 22.15.0 (v20 no longer qualifies); v25.x has zstd decompression bugs that crash during large Portal streams. See [ENVIRONMENTSETUP.md](references/ENVIRONMENTSETUP.md).

CLI: @subsquid/[email protected]. For CLI-supported EVM and SVM projects, always use programmatic mode via --config '{...}' (a JSON string or a path to a config file) instead of writing the generated scaffold by hand. Tron, Bitcoin, and Hyperliquid have no CLI templates; use the reviewed manual patterns linked below.

npm dist-tags (verified 2026-09-03): the SDK is in 1.0.0 beta. @subsquid/pipes@latest is 1.0.0-beta.6 and @subsquid/pipes@beta is 1.0.0-beta.4, so latest is ahead of beta; the CLI-generated ^1.0.0-beta.1 range therefore resolves to beta.6. Pin the exact version you tested against. The CLI trap remains: @subsquid/pipes-cli@latest is still the old 1.0.0-alpha.1; keep the explicit [email protected] pin (or use its beta dist-tag). Recheck tags with npm view @subsquid/pipes dist-tags --json before changing pins.

Beta hard-renamed the SDK surface. Current names include evmEventDecoder (was evmDecoder), mockEvmPortalStream (was evmPortalMockStream), chunkForInsert (was batchForInsert/chunk), and addRequest query-builder methods (for example, addLogRequest and addInstructionRequest). The evmPortalSource/solanaPortalSource/hyperliquidFillsPortalSource aliases are gone; only the PortalStream names exist. Transform a raw query with query.build().pipe(...) and pass the result as outputs; source-level .pipe() is gone. Projects scaffolded by the old alpha CLI pinned the floating "alpha" dist-tag, which now resolves to 1.0.0-alpha.22, so an old project can break on a fresh install. Use the full rename map in [SDKFEATURES.md](references/SDKFEATURES.md#renamed-in-the-beta-line), or pin the exact old SDK version temporarily.

Known CLI Quirks

1. Unknown table 'pipes.sync' on first run — harmless. SDK creates the table and continues.

Network Coverage

Manual setup patterns for Tron, Bitcoin, and Hyperliquid are in [TEMPLATES.md](references/TEMPLATES.md); stream/query APIs in [SDKFEATURES.md](references/SDKFEATURES.md).

Scaffolding an Indexer

See [TEMPLATES.md](references/TEMPLATES.md) for the full catalog: erc20Transfers, uniswapV3Swaps, custom for EVM; tokenBalances, custom for Solana (networkType: "svm"); manual setup for Tron, Bitcoin, and Hyperliquid.

Step 0: Research Protocol (MANDATORY)

Before writing any code, run through [RESEARCHCHECKLIST.md](references/RESEARCHCHECKLIST.md):

• Identify which contract emits the target events
• Check for proxy contracts — #1 failure mode; ~6 of 9 real indexers need manual proxy resolution
• Find the deployment block (for full history) or pick a recent start block (for faster tests)
• Decide on sink (ClickHouse recommended, PostgreSQL with Drizzle, BigQuery — see [BIGQUERYTARGET.md](references/BIGQUERYTARGET.md), Parquet, or Pub/Sub)
• Name the project

Step 1: Inspect templates (optional)

pnpx @subsquid/[email protected] init --schema

Shows all template IDs (camelCase!), required params, and target configs. The schema is also hosted at https://cdn.subsquid.io/schemas/pipescliconfig.json.

Step 2: Generate the project

pnpx @subsquid/[email protected] init --config '{
  "projectFolder": "/path/to/my-indexer",
  "packageManager": "npm",
  "networkType": "evm",
  "defaultNetwork": "ethereum-mainnet",
  "templates": [{"templateId": "erc20Transfers", "params": {"deployments": [{"address": "0xA0b86991c6218b36c1d19D4a2e9Eb0cE3606eB48", "range": {"from": "6082465"}}]}}],
  "target": "clickhouse"
}'

Template IDs must be camelCase: uniswapV3Swaps (not uniswap-v3-swaps), erc20Transfers (not erc20-transfers).

Config keys changed in the beta CLI (schema is strict — old keys are rejected): network → defaultNetwork, sink → target, and erc20Transfers/custom params are now deployment-based (deployments: [{address, range}] instead of contractAddresses + top-level range). An optional pipeId pins the generated pipe's cursor key; it is persisted to pipes.config.json so regenerating reuses the same cursor. See [TEMPLATES.md](references/TEMPLATES.md) for per-template params.

Step 3: Post-generation checklist

Proxy check is #1 priority — see [ABIGUIDE.md](references/ABIGUIDE.md) for full proxy handling.

1. Proxy contract check (custom template):

``bash grep "export const events" <project>/src/contracts/*.ts ` If only Upgraded, it's a proxy. Regenerate types from the implementation and update the import in src/index.ts — but keep the proxy address in contracts:`.

1. Project-specific database (MANDATORY):

CLI defaults the database to pipes. On @subsquid/pipes ≤ alpha.14, two indexers sharing pipes collide on the sync cursor (second one resumes from the first's position). alpha.15+ keys the cursor by pipe id, but identically named data tables (e.g. two erc20transfers) still collide — keep one database per project. ``bash DBNAME=$(basename <project-folder> | tr '-' '') docker exec <container> clickhouse-client --password <pw> \ --query "CREATE DATABASE IF NOT EXISTS $DBNAME" sed -i '' "s/CLICKHOUSEDATABASE=.*/CLICKHOUSEDATABASE=$DB_NAME/" <project-folder>/.env ``

1. ClickHouse password matches container:

``bash grep CLICKHOUSE_PASSWORD <project-folder>/.env # For an existing standalone container: match the container's password # For the generated docker-compose: "password" is correct ``

1. Contract addresses present (custom template):

``bash grep "contracts:" <project>/src/index.ts ``

1. Know your table names (custom template): one table per event, named {contractName}{eventName} in snakecase. There is no combined table.

Step 4: Start and validate

cd <project-folder>
npm run dev

Verify the first log line shows your intended start block. If it says Resuming from X, decide whether resume is correct:

Current targets key cursor rows by the source pipe id (or an explicit target settings.id). On a shared database, never drop the whole sync table to reset one pipe. Inspect it first with SELECT id, current, finalized FROM <db>.sync, confirm the id, then delete only that row. Reserve DROP TABLE <db>.sync for a database proven to be exclusive to the one pipe being reset.

Full deployment (local Docker, ClickHouse Cloud, Railway) in [DEPLOYMENT.md](references/DEPLOYMENT.md).

Troubleshooting

Match the user's symptom to a pattern. Full diagnostics in [TROUBLESHOOTING.md](references/TROUBLESHOOTING.md).

Standard diagnostic flow: read error → match pattern → read src/index.ts, package.json, .env → apply fix → restart → validate data.

Key SDK Patterns

The Pipes SDK is feature-rich — a handful of patterns cover 80% of use cases.

• Event parameter filtering (server-side): filter by indexed params at the decoder for max throughput — see [PATTERNS.md](references/PATTERNS.md#6-event-parameter-filtering-server-side).
• Factory pattern: track dynamically deployed children (Uniswap pools, MetaMorpho vaults). Includes SQLite cache, cold-start delay, d.factory?.event.* metadata — see [PATTERNS.md](references/PATTERNS.md#3-factory-pattern-with-pre-indexing).
• Topic0-only filtering: track events across ALL contracts that emit a specific signature, no address list needed. Best for protocol-unique events — see [PATTERNS.md](references/PATTERNS.md#4-topic0-only-global-filtering).
• Multi-output decoders: run multiple named decoders in one pipeline via outputs: { transfers: ..., swaps: ... } — see [PATTERNS.md](references/PATTERNS.md#5-parallel-event-decoding-multi-output).
• SDK 1.0 features: time-based ranges, defineAbi, typed errors, testing library — see [SDKFEATURES.md](references/SDKFEATURES.md).
• Tron & Bitcoin streams: native query builders and portal streams for tron-mainnet and bitcoin-mainnet — see [SDKFEATURES.md](references/SDKFEATURES.md#tron-portal-streams).
• BigQuery, Parquet & Pub/Sub targets: bigqueryTarget (partitioned tables, fork-safe DELETEs), parquetTarget (finalized-only rotating files), and pubsubTarget (BigQuery CDC rows or application-defined signals with fork handling). See [SDKFEATURES.md](references/SDKFEATURES.md#target-configuration); for BigQuery in depth, [BIGQUERYTARGET.md](references/BIGQUERYTARGET.md).
• Pipe-id-keyed cursors (alpha.15+): targets key sync state by the pipe id, so multiple pipes can share one database. Legacy ClickHouse cursors migrate automatically — see [SDKFEATURES.md](references/SDKFEATURES.md#cursor-keying--upgrading-to-alpha15).
• Beta rename map: current names are evmEventDecoder, evmPortalStream, solanaPortalStream, hyperliquidFillsPortalStream, mockEvmPortalStream, chunkForInsert. Code from alpha-era scaffolds may still say evmDecoder or evmPortalSource (a removed alias of evmPortalStream) — see [SDKFEATURES.md](references/SDKFEATURES.md#renamed-in-the-beta-line) before "fixing" either spelling.
• Portal response cache: portalSqliteCache (from @subsquid/pipes/portal-cache/node) caches Portal stream responses on disk (SQLite + zstd) to speed up re-runs and backfills over the same range; wire it via the stream's cache option — see [SDKFEATURES.md](references/SDKFEATURES.md).
• Observability: Prometheus metrics via metricsServer() (@subsquid/pipes/metrics/node, stream metrics option) and OpenTelemetry tracing via opentelemetryProfiler() (@subsquid/pipes/opentelemetry, stream profiler option) — see [SDKFEATURES.md](references/SDKFEATURES.md). For a local dashboard (progress, speed, memory, query details — multiple indexers in one view), see Pipes UI.

DeFi Protocol Forks

Many DeFi protocols share ABIs across forks — reuse saves time:

• Aave V3 forks (same Pool ABI): SparkLend, Radiant, Seamless, Granary
• Uniswap V2 forks (same Pair/Factory): SushiSwap, PancakeSwap, TraderJoe, Camelot
• Compound V2 forks (same cToken): Venus, Benqi, Tectonic

Detect by matching topic0 hashes or checking "forked from" in protocol docs.

Performance

Full benchmarks and tuning in [PERFORMANCE.md](references/PERFORMANCE.md).

Sync speed factors:

• Block range: 1M blocks ≈ 5–10 min, 5M ≈ 30–60 min, full chain ≈ 2–4 hours
• Filtering: contract events (fastest) > factory (medium) > address (slowest)
• Contract count: fewer = faster

Quick testing strategy:

1. Resolve the intended recent date and pin it as an ISO date in range.from (for example, the date seven days before the test)
2. Limit to 1–3 contracts first
3. Expand once working

Data Validation

Before declaring an indexer production-ready, run the checks in [VALIDATION.md](references/VALIDATION.md):

• Table structure matches design
• No NULLs in required fields
• Addresses match ^0x[0-9a-fA-F]{40}$
• Transaction hashes are 66 chars
• Block range complete (no gaps)
• Data increases over time
• Sample rows match block explorer

Long-Running Indexers

For production indexers that must survive crashes and reboots, see [STREAMRESILIENCE.md](references/STREAMRESILIENCE.md): retry patterns, pm2 supervisor, nohup for dev sessions.

Analytics Queries

Dashboard-grade ClickHouse patterns (time bucketing, conditional aggregation, parameterized WHERE) in [CLICKHOUSEANALYTICS.md](references/CLICKHOUSEANALYTICS.md).

Reference Files

Official Docs

• llms.txt quick reference: docs.sqd.dev/llms.txt
• Full documentation: docs.sqd.dev/llms-full.txt
• Comprehensive SDK guide: docs.sqd.dev/skill.md
• Available datasets: portal.sqd.dev/datasets

Related

• portal: query blockchain data across 140+ networks via Portal MCP or the SQD Portal Stream API. Use it to verify contract events, discover dataset names, cross-check indexed data, and decide when a query should become a Pipes/Squid pipeline.
• Ponder on Portal — if the user already has a Ponder indexer, @subsquid/ponder backs its historical sync with Portal as a drop-in (same handlers/schema; one portal: line per chain) — no Pipes rewrite needed.