Changelog

All notable changes to this project are documented here. Starting with 1.0.0, releases follow Semantic Versioning.

Unreleased

Added one-command setup that accepts password-free database URLs or SQLite paths, prompts for missing details, verifies the database, and installs the managed skill in ~/.agents/skills.
Made setup reports readable text by default. --format json provides structured output for automation. Query output remains JSON by default.
Added asterisk-masked password input and separate managed credential storage. Existing native authentication remains available, with explicit environment overrides and deterministic non-interactive setup.
Added verified TLS for new setup targets, including private CAs and native client certificates. Existing targets retain their TLS settings.
Expanded config check with offline diagnostics, credential-source reporting, read-only session checks, and bounded catalog verification. Configuration inspection no longer synchronizes skills.
Preserved unrelated configuration, comments, credentials, and protected skill content during repeated setup. Password replacement verifies the new credential before switching its configuration reference.
Improved safe recovery messages, absolute SQLite paths, native connection discovery, and SQLite schema inspection with newer SQLGlot versions.
Rewrote the PowerShell and Bash quick starts and documented permissions, credential renewal, and agent execution environments. Added onboarding and packaging coverage.

Added local automatic synchronization of already-installed managed agent skills, using the running CLI version and front matter content hashes to preserve modifications and prevent downgrades.
Added skill install --force, read-only skill status, and skill remove. Existing install-skill and remove-skill commands remain supported.
Excluded local development builds and custom skill locations from automatic synchronization. Skill updates use atomic replacement and stderr notices without changing command results.
Kept unrelated files when removing a skill and prevented installation from overwriting unmanaged content.

Added config check for safe, structured connectivity diagnostics.
Added --about package metadata.
Added real PostgreSQL and MySQL integration coverage and broader Python/Windows CI.
Normalized driver failures into concise exit-code-1 diagnostics without tracebacks or exposed passwords.
Added read-only PostgreSQL and MySQL/MariaDB session settings and direct integration checks that sessions reject writes.
Added a SQLite execution timeout while retaining read-only file access.
Hardened SQL validation against mutating CTEs, locking reads, unsafe PostgreSQL file functions, stacked statements, and non-allowlisted SQLite pragmas.
Made TLS modes explicit: required fails without negotiated encryption, preferred may fall back to plaintext, and disabled prohibits TLS.
Added and refined the managed $sql-agent-cli skill, including uvx invocation, troubleshooting, least-privilege, and TLS guidance.
Documented the public JSON, config, stream, exit-code, semantic-versioning, deprecation, and security contracts targeted for 1.0.