Polygres Agent Skills 0.5.0

Added

• Added synchronized-project guidance across all five skills for Supabase, Neon, and PostgreSQL sources.
• Added deterministic pipeline-plan guards for synchronized projects, including target schema, imports, writes, credentials, capture workers, and sync-control boundaries.
• Added sync-aware scaffolding and review output for source authority, table selection, managed replication, reconfiguration, and source-only writes.
• Added preflight, lifecycle, generation-conflict, schema-drift, resync, and permission troubleshooting guidance.

Changed

• Required project-mode resolution before choosing CLI, SDK, database, ingestion, or retrieval surfaces.
• Directed synchronized-project creation and initial table selection to the CLI or dashboard while keeping source credentials out of agent-visible paths.
• Documented the standard and synchronized project creation command surfaces.
• Restricted synchronized-project SDK and API-key guidance to supported retrieval, catalog, and readiness surfaces.
• Updated Context and retrieval design for embeddings generated in synchronized source databases.
• Coordinated guidance with polygres-sdk 0.4.0 and polygres-cli 0.4.0.

Polygres Agent Skills 0.4.0

Added

• Added polygres-data-pipeline for guided ingestion and retrieval setup from
  conversations, databases, files, APIs, event streams, and existing Polygres
  data.
• Added read-only device inspection and user choice for local Ollama, Sentence
  Transformers, llama.cpp, and application-owned ONNX embedding options.
• Added a versioned pipeline-plan validator, setup-pack scaffolder, safe .env
  status checker, local embedding adapters, and an idempotent SQLite checkpoint
  ledger.
• Added separate documented store and retrieve interface decisions for CLI,
  SDK, Runtime API, and approved direct Postgres fallback workflows.
• Added a versioned local and hosted embedding model catalog plus deterministic,
  device-aware recommendation tooling.

Changed

• Expanded the plugin from four skills to five.
• Added one-line setup prompts such as Look at my data and use $polygres-data-pipeline to set up a Polygres data pipeline.
• Required generated local setup packs to ignore .env, keep credential values
  out of agent context, and give the user exact local paste instructions.
• Changed pipeline setup to a bounded build-first fast path. The skill creates
  and tests source-specific runtime code before full source inspection,
  backfill, or broad Polygres capability discovery.
• Limited questions to genuine blockers and consolidated upload and remote
  mutation approval into one review.
• Updated CLI and SDK guidance for the single-row Runtime write surface,
  including exact no-retry behavior and ambiguous outcomes.
• Made Context-backed capture one user-facing row operation that also completes
  or durably starts point reconciliation.
• Defined per-surface checkpoint state for pending or partially failed Context
  reconciliation, with exact idempotent replay instead of a manual point step.
• Preserved generic row-only behavior unless a Context collection is explicitly
  selected or resolved safely.
• Moved an unknown local-versus-hosted embedding choice into the single setup
  review, so selecting either fully disclosed option does not require another
  approval.
• Clarified that workflow order, numeric values, retrieval timing, capture,
  graph, embeddings, and agent integration are adaptable guidance rather than
  required pipeline components.
• Made generated setup packs omit explicitly disabled components and allowed
  managed agent instructions to configure capture, recall, or both.
• Made fully vague setup requests ask one short source-and-outcome question
  before inspection, while contextual and detailed requests keep the fast path.
• Added read-only personalized project recommendations for questions such as
  What can I do with Polygres? without starting setup automatically.
• Added a direct reply that carries an accepted recommendation into setup
  without repeating discovery or bypassing the mutation review.

Polygres Agent Skills 0.3.0

Added

• Added pgContext guidance across the CLI, SDK, retrieval-design, and troubleshooting skills.
• Added collection lifecycle, point ingestion, search, migration, and durable-operation workflows.
• Added guidance for selecting relational, graph, text, hybrid, pgvector, and pgContext strategies.

Changed

• Prefer AI Context for new semantic-search setups while preserving pgvector compatibility guidance.
• Route retired vector-config workflows to supported Context commands and SDK methods.
• Expanded version, installation-origin, source-install, pgGraph fixture, and RLS verification guidance.

Polygres Agent Skills 0.2.0

Highlights

• Adds polygres-retrieval-design for reviewable relational, graph, vector, text, hybrid, and grounded RAG plans without direct mutation.
• Adds polygres-troubleshooting for read-only diagnosis across CLI, control-plane, Runtime API, Postgres, jobs, migrations, and retrieval.
• Retains the existing polygres-cli and polygres-sdk skills.
• Updates Codex and Claude plugin metadata and public documentation for the four-skill package.

Compatibility

• Skill/plugin version: 0.2.0
• Declared minimum polygres-cli version: 0.1.0
• Maximum contract-tested CLI manifest version: 0.2.0
• Declared minimum polygres-sdk version: 0.1.0
• Codex CLI tested: 0.144.2
• Claude Code tested: 2.1.81
• skills CLI tested: 1.5.17

Verification

• 51 fast agent-skills tests and 6 heavy contract tests passed.
• 92 related Python CLI tests and 16 Python SDK tests passed.
• All four skills, the Codex plugin, and the Claude marketplace passed their official validators.
• Deterministic multi-skill export comparison passed.
• A clean project-scoped install from Evokoa/polygres-skills discovered and installed all four skills.
• Retrieval-design and troubleshooting passed isolated read-only forward evaluations.

Content digest

SHA-256 of git archive for commit 5a6839d:

7e403ea448b8418ac2282d9f6e1bfbd2dc4c56ee7b043aa1bec705e879a306a5

Known limitations

• No new CLI commands are included because no additional released public CLI contract was available.
• Database workflows and native skills install/status/uninstall remain gated by unfinalized public execution contracts.
• Organization guidance remains in product documentation rather than a standalone skill.
• Automatic activation, native marketplace visual review, and live mutating evaluations against disposable projects were not run.