Polygres Agent Skills connect coding agents to Polygres MCP, guide PostgreSQL sync and data-pipeline setup, operate projects, design retrieval, build applications, and diagnose failures through supported Polygres interfaces.
User guide: Polygres Agent Skills
| Skill | Use it for |
|---|---|
polygres-data-pipeline |
Use available MCP tools to choose managed PostgreSQL sync or build ingestion, retrieval, memory, or agent integration. |
polygres-cli |
Use MCP for compatible interactive operations and the CLI for authentication, migrations, keys, imports, and retrieval configuration. |
polygres-sdk |
Use MCP for grounded interactive retrieval and build persistent Python integrations. |
polygres-retrieval-design |
Inspect read-only MCP evidence and produce a retrieval implementation plan. |
polygres-troubleshooting |
Diagnose MCP, CLI, API, PostgreSQL, job, migration, and retrieval failures using read-only evidence. |
Compatible agents select the appropriate skill automatically. You can also name the skill in your request when you want a specific workflow.
npx skills add Evokoa/polygres-skillsTo install globally for Codex and Claude Code:
npx skills add Evokoa/polygres-skills \
--global \
--agent codex \
--agent claude-code \
--yescodex plugin marketplace add Evokoa/polygres-skills
codexOpen /plugins, choose the Polygres marketplace, install Polygres, and start a new task.
The Codex plugin installs the production Polygres MCP connection and all five skills. Its base connection covers all accessible projects. Use the Polygres Dashboard's project Connect → MCP page when you want a fixed-project connection.
Run these commands inside Claude Code:
/plugin marketplace add Evokoa/polygres-skills
/plugin install polygres@polygres
/reload-plugins
Ask for the outcome you want in one short line or a detailed specification. The skill infers the workflow from intent, inspects relevant state, and asks only for critical information it cannot safely discover.
Use Polygres MCP to inspect this project and recommend the next useful setup step.
Help me set up Polygres.
What can I do with Polygres?
Look at my data and use $polygres-data-pipeline to set up a Polygres data pipeline.
For a request that names a source or outcome, the pipeline skill takes a
bounded sample, applies safe defaults, selects only useful components, and
creates runnable source-specific code. For a fully vague request such as Help me set up Polygres, it first asks one short question about the desired outcome
and relevant data, then begins inspection. Schema changes, embeddings, graph,
backfill, continuous capture, retrieval, and agent instructions are optional.
Sample sizes, retrieval timing,
and numeric limits are starting defaults that adapt to the user's setup. Before
remote mutation or changing active agent instructions, it shows one concise
review and asks once. Internal lint warnings do not become user questions.
For What can I do with Polygres?, the skill performs a bounded, read-only
scan of the accessible workspace and current Polygres project, then recommends
the most useful next step for that specific project. It does not change or
scaffold anything until the user chooses a recommendation. The response ends
with a direct setup reply, such as Set up the recommended Polygres pipeline.
That reply carries the recommendation into the setup flow without repeating
discovery; the skill still shows the consolidated review before changes.
The same adaptive flow works with prompts such as:
Look at my conversations and set up a Polygres data pipeline.
Look at my current setup and set up a Polygres data pipeline.
Sync my Supabase, Neon, or PostgreSQL database into Polygres.
Log me into Polygres and help me select the correct project.
Import customers.json into public.customers. Inspect it first and explain any
conversion choices before changing data.
Configure Polygres AI Context retrieval for documents.embedding with 1536
dimensions and verify readiness.
Use the Polygres SDK to retrieve similar documents, expand their citations,
and build deduplicated context with source references.
Design a retrieval plan for this schema. Compare relational, graph, text,
hybrid, Polygres AI Context, and any existing vector configuration without
changing the project.
Diagnose why this pgContext collection is blocked. Use read-only evidence and
recommend the safest next action.
The skills follow a few important boundaries:
- They use public Polygres CLI, Runtime API, SDK, and PostgreSQL interfaces.
- They resolve project mode before choosing a surface. Synced projects keep the
source database authoritative and reject target rows, imports, migrations,
SQL, database credentials, and
psql. - They hand synced-project creation, source preflight, table selection, and lifecycle work to the dashboard. Source credentials never enter agent chat, generated plans, CLI arguments, Runtime requests, or SDK code.
- The pipeline skill records separate documented store and retrieve interfaces, selecting CLI, SDK, or Runtime API based on the workload. CLI and SDK 0.4.0 provide capability-gated single-row validation, insert, upsert, and ignore. On standard projects, direct Postgres remains an explicitly approved compatibility fallback.
- They ask before imports, migrations, revocations, deletions, and schema changes.
- They keep database passwords out of command arguments and generated code.
- They treat Runtime API keys as secrets and warn when a command can expose one in terminal or agent history.
- They keep authorization in the application. Retrieval filters can narrow results, but they do not replace access control.
- They preserve request IDs and relevant resource IDs when diagnosing a failure.
- They honor a known embedding preference. If local versus hosted is unknown, they silently inspect compatibility and include one local recommendation and one hosted alternative in the single setup review. Selecting either reviewed option is the one approval. Polygres does not generate embeddings.
- They generate
.env.example, ensure.envis ignored, and tell the user how to paste credential values into.envwithout exposing them to the agent.
The Polygres CLI imports CSV directly. The CLI skill can safely prepare TSV, JSON arrays, and JSONL or NDJSON as CSV before starting an import. It does not upload the original source file.
Export Excel, Parquet, Avro, ORC, XML, YAML, SQL dump, and custom pg_dump sources to CSV or JSONL before using this workflow.
Update an Agent Skills installation:
npx skills update polygres-data-pipeline
npx skills update polygres-cli
npx skills update polygres-sdk
npx skills update polygres-retrieval-design
npx skills update polygres-troubleshootingRefresh the Codex marketplace:
codex plugin marketplace upgrade polygresThen open /plugins to update or reinstall Polygres if prompted.
For Claude Code:
/plugin marketplace update polygres
/plugin update polygres@polygres
/reload-plugins
Remove a global Agent Skills installation:
npx skills remove --global polygres-data-pipeline
npx skills remove --global polygres-cli
npx skills remove --global polygres-sdk
npx skills remove --global polygres-retrieval-design
npx skills remove --global polygres-troubleshootingFor Codex, uninstall Polygres through /plugins, then optionally remove the marketplace:
codex plugin marketplace remove polygresFor Claude Code:
/plugin uninstall polygres@polygres
/plugin marketplace remove polygres
/reload-plugins
Package version: 0.6.0. It supports Polygres MCP catalog 1.0, polygres-cli 0.4.0 through 0.4.1, and polygres-sdk 0.4.0 through 0.4.1. If an example differs from your installed version, follow discovered MCP tools, installed CLI help, or the SDK method signature.
See the Agent Skills 0.6.0 release notes for release changes.
Apache License 2.0. See LICENSE.