Agent Skills for operating managed Polygres projects, designing retrieval, building Python applications, and diagnosing failures through public surfaces.
The plugin contains four independently triggered skills:
polygres-clifor project operations, imports, migrations, keys, and retrieval configuration;polygres-sdkfor graph, vector, text, hybrid, and grounded RAG application code;polygres-retrieval-designfor reviewable relational, graph, vector, text, hybrid, and RAG plans without mutating a project;polygres-troubleshootingfor read-only diagnosis using public CLI and SDK evidence.
Source repository: Evokoa/polygres-skills
Install with the cross-agent skills installer:
npx skills add Evokoa/polygres-skillsInstall explicitly for Codex and Claude Code at user scope:
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, and install Polygres. Start a
new task after installation.
Inside Claude Code:
/plugin marketplace add Evokoa/polygres-skills
/plugin install polygres@polygres
/reload-plugins
Compatible agents can activate the appropriate skill automatically from the request. Invoke a skill explicitly when you need to select the CLI or SDK workflow yourself:
/polygres:polygres-cli
/polygres:polygres-sdk
/polygres:polygres-retrieval-design
/polygres:polygres-troubleshooting
Update a generic Agent Skills installation:
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 generic installation:
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
Describe the outcome you want. The agent selects the documented CLI workflow and asks before destructive or secret-producing operations.
Example prompts:
Log me into Polygres and help me select the correct project.
Import customers.json into public.customers. Inspect it first, explain any
conversion choices, and ask before changing data.
Review this SQL migration and ask before applying it to my selected Polygres
project.
Configure vector retrieval for documents.embedding with 1536 dimensions and
verify readiness.
Use the Polygres SDK to retrieve semantically similar documents, expand their
citations, and build a deduplicated context with provenance.
Design a reviewable retrieval plan for this schema. Compare relational, graph,
vector, text, and hybrid strategies without changing the project.
Diagnose why this import timed out. Preserve job and request IDs, use only
public read-only evidence, and recommend the safest next action.
Use polygres-retrieval-design before configuration when the retrieval model
or strategy is unsettled. It records verified schema facts, unresolved
assumptions, graph and embedding choices, authorization, provenance,
deduplication, token budgets, validation, and separate CLI and SDK handoffs.
Use polygres-troubleshooting after a failure. It resolves identity and exact
project context, checks public readiness and job state, distinguishes local
CLI, control-plane, Runtime API, and Postgres failures, and retains request and
job IDs. It does not use private observability or perform repair mutations.
Install polygres-sdk in the application environment separately from
polygres-cli. The SDK skill helps select the project Runtime API URL, load the
API key from server-side environment configuration, check readiness, and write
typed graph, vector, text, or hybrid retrieval code.
It preserves these boundaries:
- application retrieval uses the per-project Runtime API, not the central control-plane or Postgres URLs;
- graph calls use real row IDs returned from trusted application data or prior retrieval results;
- connection information is passwordless;
- filters narrow retrieval but do not replace application authorization;
- pagination, request IDs, provenance, deduplication, and context token budgets remain explicit.
The current Polygres CLI imports CSV. The skill can prepare TSV, JSON arrays, and JSONL/NDJSON locally as CSV before invoking the supported import command. The preparation workflow does not upload the original non-CSV file and does not make network requests.
Excel, Parquet, Avro, ORC, XML, YAML, SQL dump, and custom pg_dump conversion
are outside the first release. Export those sources to CSV or JSONL first.
- Browser login is the public authentication flow.
- Database passwords are never retrieved or passed in command arguments.
- Runtime API-key secrets are shown once and can enter terminal or agent history. Run key creation directly when transcript exposure is unacceptable.
- Replacement imports, migrations, revocations, deletes, and schema mutations require explicit approval.
- The skill uses the public CLI and does not call private Polygres endpoints.
Skill release 0.2.0 targets polygres-cli >= 0.1.0 and polygres-sdk >= 0.1.0. Installed CLI help and the installed Python SDK signatures remain
authoritative when local package versions differ from examples.
Apache License 2.0. See LICENSE.