Changelog

Unreleased

[0.3.0] - 2026-10-29

Runtime support remains unverified under the owner's waiver; static checks do not establish client compatibility.

Removed

• The agent-team package excludes the changelog, open-issues, and project-status rules so project-record conventions can be installed independently.

Fixed

• The package README distinguishes the current manifest from tagged releases while retaining the unverified runtime warning.

[0.2.0] - 2026-10-28

Changed

• The agent-team package replaces copilot-workbench in the awieczork-tools marketplace with the Agent Plugins 1.0 layout and preserved team and database behavior.
• Agent Host loading of the namespaced agents and rules remains unverified under the owner's runtime-verification waiver.

Removed

• The agent-team package excludes copilot-customization; the separate agent-plugin-dev package supplies the skill.

Fixed

• The agent-team README identifies the migration package, states the unverified runtime risk, and gates installation and updates on authorized publication.

[0.1.3] - 2026-10-16

Added

• Continue a discussion in another session. Invoke seed-task to turn the part of a conversation you choose into a task you can start separately. You confirm what it carries, then receive a start prompt so the next session has the context it needs.

Changed

• Use one lead for workspace changes. Select lead, which replaces software-lead, for code, documentation, configuration, or a change that needs data evidence. It coordinates specialists, runs independent work in parallel, and checks that the parts work together before delivering the result.
• Resume planned work after an interruption. When lead or data-lead uses a plan, it saves the accepted plan with its amendments and completed tasks. A resumed session follows that current plan rather than reconstructing progress from the conversation.
• See what supports a data answer. Use data-lead for work that uses or depends on data. Analysis results include row counts, filter checks, and coverage; before an analysis supports a decision, the lead also checks its framing and has key figures recomputed independently.
• Develop an idea and its documentation in one discussion. Ask discuss to investigate an idea, create or edit documentation, and shape a request for a lead. Use seed-task when you want to carry the discussion into a new session instead of continuing through a handoff.
• Review a draft before settling customization choices. When the scope or type of an agent customization is unclear, copilot-customization shows proposed content before asking you to choose, so the decision rests on a concrete draft.
• Use the Redshift fallback without confirming its target by hand. The fallback reports the host and database it used, and results from another target are rejected. Scripts that call the runner must handle exit status 1 for input, connection, and query errors instead of the former separate error codes.

[0.1.2] - 2026-10-15

Fixed

• The plugin uses the Copilot plugin format for agent discovery in Copilot (Agent Host) sessions and agent and rule discovery in the Agent Customizations editor.

[0.1.1] - 2026-10-15

Added

• The awieczork-tools marketplace provides installation and updates for copilot-workbench through VS Code's /plugin commands.
• The README explains how to set CONTEXT7_API_KEY in the MCP server's environment for authenticated Context7 access.

Changed

• The documented VS Code setup uses the Copilot harness in Agent Host.

[0.1.0] - 2026-10-10

Added

• The software-lead agent delivers or plans software changes by deciding the approach, briefing workers, and judging their results.
• The data-lead agent answers data-science questions with evidence, explicit uncertainty, and decisions left to the user.
• The discuss agent shapes an unfinished idea into a request for a lead or a self-contained prompt for a new session.
• The explorer worker locates workspace evidence for one question.
• The database-explorer worker discovers and profiles SQL Server and Redshift objects for one data question, verifies join paths, and returns read-only queries.
• The data-explorer worker profiles local datasets, their joins, and their splits without changing source files.
• The researcher worker gathers sourced external evidence with its support, conflicts, and gaps.
• The planner worker produces executable plans for software changes and data analyses, and returns open decisions to the delegating agent.
• The developer worker implements software changes with tests scaled to the cost of an undetected error.
• The ui-developer worker implements interface changes and checks them in the running app in a browser.
• The reviewer worker assesses finished results against their goal without repairing them.
• The challenger worker tests one decision or framing and recommends whether to proceed, revise, or reject it.
• The copilot-customization skill guides choosing, writing, and reviewing Copilot project instructions, scoped instructions, custom agents, and Agent Skills for VS Code.
• The brief-subagent skill supplies self-contained, one-shot briefs and how to act on the status each worker returns.
• The task-context skill keeps a task's summary and delegated results in a Git-ignored folder, so that a task survives the session and resumes later.
• The user-invoked interview-me skill confirms the user's intent before work continues.
• The redshift skill runs read-only queries against user-provided targets with an MCP-first workflow and a Python fallback that returns at most 100 rows.
• The sql-server skill runs read-only queries against user-provided targets through verified saved connections.
• The run-python skill runs Python scripts and inline code through uv, with or without the project environment.
• The markdown rule supplies document syntax, formatting, and prose rules, and only its syntax rules for customization files.
• The python rule supplies conventions for user-run task scripts with proportionate validation.
• The sql rule supplies schema reading, result bounding, and query text rules for read-only SQL Server and Redshift queries.
• The changelog rule keeps a project's CHANGELOG.md in Keep a Changelog sections.
• The project-status rule keeps a project's PROJECT-STATUS.md as a short handoff with its roadmap and blockers.
• The open-issues rule records and closes a project's unresolved decisions, missing inputs, and verified limitations in OPEN-ISSUES.md.
• The Redshift MCP server runs database queries and catalog discovery with local AWS configuration.
• The context7 MCP server provides current library documentation.