Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog.

2026-07-28​

Changed​

• Agent model arrays — Reconciled and synchronized the audited agent model: arrays and cross-provider fallbacks against the required target order; Gemini 3.1 Pro (Preview) remains unchanged only for the BA analysis worker's separate agent-level assignment.
• Gemini update — The confirmed active Gemini 3.5 Flash references (tsh-software-engineer, the Engineering Manager routing prose, and the orchestration routing table) now use Gemini 3.6 Flash, including the Software Engineer website documentation.
• Prompt-authoring guidance — tsh-creating-prompts skill and template, and the corresponding website documentation (Prompts Overview, Creating Prompts, /tsh-create-custom-prompt, Copilot Engineer), are aligned so public prompts infer their model from the bound agent: field and omit a prompt-level model: field; exactly the nine scoped public prompts (including tsh-review-ui) had their prompt-level model: declarations removed, and public prompts now infer model selection from agent:.

2026-07-27​

Changed​

• Human Approval precondition — Normalized one structurally uniform inline <human-approval-precondition> block across all seven execution-owner agents, retaining the sanctioned UI scope qualifier only in tsh-ui-engineer.
• Execution owners — Fail closed when the referenced plan cannot be located or read, after one bounded resolution attempt.
• Guided recovery — Owners name the exact failed field, condition, or file and use vscode/askQuestions on every entry path; hand-back to tsh-engineering-manager is one offered choice for delegated subagents.
• Agent tools — Added vscode/askQuestions to the tsh-devops-engineer and tsh-technical-writer tool lists.
• tsh-orchestrating-implementation — Now owns route-varying authorization bases, eligibility, and escalation, while the common invariant stays inline in execution owners.
• Role boundaries — Clarified that tsh-engineering-manager and tsh-architect present and record Human Approval while execution owners validate before editing; tsh-plan-reviewer remains non-implementing.
• Portable harness documentation — Made documentation-target wording conditional on those targets existing in the project.

2026-07-22​

Changed​

• tsh-orchestrating-implementation — Added an exact-revision Human approval gate, keeping automated Reviewer approval distinct from Human approval.
• Implementation routing — Missing research or plan now routes to preparation rather than planless execution.
• tsh-creating-implementation-plans — Added a persisted Human Approval revision record to the plan-state schema and template.
• Routed execution owners — Added a defense-in-depth pre-file-change predicate to all seven routed execution owners.
• Public prompt and documentation — Synchronized the public implementation prompt and documentation with the Human approval gate, preparation routing, and execution-owner preconditions.

2026-07-10​

Changed​

• Model arrays updated across Copilot customization agents and public prompts:
  • tsh-architect, tsh-business-analyst, tsh-context-engineer, tsh-prompt-engineer, and tsh-ba-quality-worker now use GPT-5.6 Terra with GPT-5.4.
  • tsh-ba-formatting-worker, tsh-ba-transcript-worker, tsh-copilot-artifact-creator, and tsh-technical-writer now use GPT-5.6 Luna with GPT-5.4 mini.
  • tsh-copilot-artifact-reviewer and tsh-plan-reviewer now use GPT-5.6 Sol with GPT-5.5.
  • tsh-copilot-orchestrator now uses GPT-5.6 Terra with Claude Sonnet 5; tsh-engineering-manager and /tsh-implement now use GPT-5.6 Luna with Claude Sonnet 5.
  • Public /tsh-analyze-aws-costs, /tsh-analyze-gcp-costs, /tsh-analyze-materials, /tsh-explore-materials, /tsh-audit-infrastructure, and /tsh-review-codebase prompts now use GPT-5.6 Terra with GPT-5.4.
  • tsh-software-engineer retains Kimi K2.7 Code, GPT-5.3-Codex, and Gemini 3.5 Flash; tsh-plan-implementor retains qwen3-coder-30b-a3b-instruct (customendpoint), MAI-Code-1-Flash, and GPT-5.4 mini.
• Website docs — Synced the Engineering Manager agent page, Software Engineer agent page, and /tsh-implement prompt page to reflect the current model arrays rather than a Claude-only model.
• README guidance — Added recommendations for manually configuring thinking effort for each model in the VS Code Copilot Chat UI; the configuration becomes active for the current VS Code profile:
  • GPT-5.6 Sol — medium (default)
  • GPT-5.6 Terra — medium/high
  • GPT-5.6 Luna — medium/high
  • Sonnet 5 — high (default)
  • MAI-Code-1-Flash — high

2026-06-22​

Added​

• tsh-ui-engineer agent — New UI-specialized implementor (model Claude Sonnet 4.6) that owns frontend and user-interface implementation. Carries the full UI toolset (Figma, Playwright, Context7, Sequential Thinking), the frontend skill bundle (tsh-implementing-frontend, tsh-implementing-forms, tsh-writing-hooks, tsh-ensuring-accessibility, tsh-optimizing-frontend, tsh-ui-verifying), delegates verification to tsh-ui-reviewer, and confirms scope via vscode/askQuestions before proceeding without a plan.
• tsh-plan-implementor agent — New internal-only (user-invocable: false) strict implementor that executes one plan task at a time exactly as written, with a minimal toolset and a stop-and-report path for missing seams or ambiguous plans. Reuses the shared tsh-implement-common-task.prompt.md rather than adding a dedicated prompt; first in-repo use of a multi-model array with a customendpoint availability fallback.
• Documentation pages for the UI Engineer and Plan Implementor agents on the website.

Changed​

• tsh-software-engineer agent — Refactored into the standard non-UI implementor: removed the Figma and Playwright tools and the UI skill bundle, dropped the tsh-ui-reviewer subagent, switched to a GPT-5.3-Codex / Gemini 3.5 Flash model array with selection guidance, and added an explicit vscode/askQuestions no-plan confirmation step. UI work now routes to tsh-ui-engineer.
• tsh-orchestrating-implementation skill — Split the single implementor route into three: UI with Figma to tsh-ui-engineer, strict low-risk plan tasks to tsh-plan-implementor, and complex/no-plan non-UI work to tsh-software-engineer; added the no-plan vscode/askQuestions confirmation gate, the software-engineer model-selection note, and excluded tsh-plan-implementor from Quick Flow.
• tsh-engineering-manager agent — Registered tsh-ui-engineer and tsh-plan-implementor in the subagents list and delegation roster, narrowed the tsh-software-engineer entry to non-UI implementation, and reconciled the constraints so no single implementor is treated as owning every implementation path.
• Reference reconciliation — Repointed UI ownership to tsh-ui-engineer across tsh-ui-reviewer, plan.example.md, and the tsh-implement-ui internal prompt, while keeping tsh-software-engineer as the default non-UI fix target in tsh-code-reviewer and tsh-e2e-engineer with an explicit UI exception.
• Website docs — Reframed the Software Engineer page as non-UI, updated the agents overview table and handoff diagram to include the three implementors, and added the new agent pages.
• Follow-up hardening — Normalized the tsh-software-engineer, tsh-ui-engineer, and tsh-plan-implementor bodies to the canonical XML-tag structure, nested plan-progress guidance inside <agent-role>, and removed the in-body software-engineer model-selection prose while keeping its frontmatter model array intact.
• Follow-up hardening — Made tsh-plan-implementor the default route for approved, actionable, low-risk plan seams across tsh-orchestrating-implementation and tsh-engineering-manager, with tsh-software-engineer as the explicit exception for complex or no-plan non-UI work; orchestration now owns delegation-time model selection between GPT-5.3-Codex and Gemini 3.5 Flash, and Quick Flow allows tsh-plan-implementor when an approved plan seam already exists.
• Follow-up hardening — Aligned the published Software Engineer docs page with the orchestrator selecting the model at delegation time and matched its wording to the hardened three-implementor routing contract.

2026-06-19​

Added​

• tsh-technical-writer agent — Internal worker agent that owns repository documentation, authoring and updating README, CHANGELOG, in-repo /docs, and published documentation-site pages. Delegated to by the Engineering Manager for documentation-only work; never writes or edits product code.
• tsh-writing-documentation skill — Canonical documentation-writing skill covering README, CHANGELOG, in-repo /docs, and the published documentation site. Includes documentation scope rules, accuracy-over-volume principle, structure-mirrors-neighbors convention, broken-link policy, and reader-centered craft guidelines from Writing for Busy Readers.
• Internal prompt tsh-write-documentation — Worker prompt that hands off a bounded documentation task to tsh-technical-writer; loads tsh-writing-documentation and tsh-technical-context-discovering skills before authoring begins.
• Documentation pages for the Technical Writer agent and Writing Documentation skill on the website.

Changed​

• tsh-engineering-manager agent — Added tsh-technical-writer to the subagents list; added delegation rules (when to delegate documentation-only work and when not to); tightened the "never writes product code" constraint to "never edits any file directly"; added read/search tool guardrails (routing decisions only, never for research or solving).
• tsh-orchestrating-implementation skill — Renamed never-writes-product-code principle to never-edits-files-directly (now covers all file types, not just product code); added read-search-routing-only and last-resort-stop-or-ask principles; added repository documentation as a first-class work type routed to tsh-technical-writer via tsh-write-documentation.prompt.md; tightened delegation handoff contract (bounded task slice, no raw context dumping); tightened fix routing rule (structured per-finding list before re-delegating).
• Website agents overview — Added Technical Writer to the Engineering Manager delegation diagram and to the Internal Worker Agents table.

2026-06-18​

Added​

• Open-questions dispatch gate for implementation plans — plans with any ❓ Open rows now block execution until tsh-architect resolves them.
• Executable-slot dispatch gate — plans can’t be dispatched if any verification field, DoD command, or file path still contains placeholder/default values.
• Planning-readiness gate in tsh-orchestrating-implementation — execution now checks that open questions are cleared before proceeding.

Changed​

• tsh-creating-implementation-plans skill — hardened the plan contract:
  • Wildly Important Goal now explicitly requires Goal, Success Measure, and Do NOT touch / do NOT add
  • phases now require a Verification field with exact fast-running checks
  • tasks now require Files: entries with create / modify / reuse labels
  • DoD now distinguishes code tasks from docs/config tasks and requires stack-specific runnable checks or deterministic file assertions
  • UI verification is clarified as distinct from full e2e
• plan.example.md template — updated the canonical template to match the stricter plan contract, including the new goal hierarchy, verification fields, docs-only task pattern, and comment-wrapped implementation note.
• tsh-review-plan.prompt.md — now blocks review when ## Open Questions still contains ❓ Open, and it requires a compact Decision and Revision History section.
• Website docs for Creating Implementation Plans — synced to the new template and workflow rules.

2026-06-15​

Added​

• tsh-creating-implementation-plans skill — centralized plan structure ownership, phase/task ordering, definition-of-done rules, and the canonical plan template/example.
• Direct architect/plan-reviewer nesting — tsh-architect can now hand off plan review directly to tsh-plan-reviewer with a strict review-report schema.

Changed​

• tsh-architect agent — tightened into a WHO-only architecture role and delegated plan structure ownership to tsh-creating-implementation-plans.
• tsh-plan-reviewer agent — improved tool-usage guidance and aligned it with the architect-owned review loop.
• tsh-orchestrating-implementation skill — now explicitly plans task order and coordinates the updated architect → plan-reviewer flow.
• tsh-plan / tsh-review-plan prompts and tsh-architecture-designing skill — removed duplicated plan-authoring rules and moved them into tsh-creating-implementation-plans.
• Engineering Manager / website docs — updated docs for the refactored architect, plan reviewer, and engineering manager flow.
• Installation guide — added the VS Code setting required to allow nested agents.

2026-06-11​

Added​

• tsh-orchestrating-implementation skill — Added the canonical implementation orchestration workflow with flow selection, delegation routing, todo control, and review gates.

Changed​

• tsh-engineering-manager agent — Reworked into a WHO-only orchestrator that delegates implementation work through tsh-orchestrating-implementation.
• /tsh-implement prompt — Reduced to a thin trigger that hands off to the orchestration skill.
• Website documentation — Updated the Engineering Manager and /tsh-implement docs to reflect the new orchestration flow.

Fixed​

• tsh-orchestrating-implementation skill — Added the missing fallback policy for unsupported tasks and the rule to record solution changes in the plan changelog.
• tsh-orchestrating-implementation skill — Fixed Quick Flow always delegating to tsh-software-engineer regardless of task domain; added a shared Task-to-Owner Routing table so Quick Flow (and its review fix-loop) routes CI/CD, infra/Terraform, Kubernetes/deploy, observability, and LLM-prompt tasks to their correct owner and prompt instead of the wrong one.

2026-06-01​

Added​

• /tsh-explore-materials prompt — Added a Business Analyst exploration mode for ambiguous workshop inputs; produces workshop-context-summary.md before backlog extraction begins

Changed​

• Business Analyst agent (tsh-business-analyst) — Reworked into an orchestrator that delegates transcript cleanup, context synthesis, extraction, quality review, and Jira formatting to internal BA workers while retaining all user-facing gates and Jira mutations
• /tsh-analyze-materials prompt — Added Gate 0 intent-brief approval, Explore Mode support, Lite/Full quality review with Gate 1.5, post-push Jira verification, and project baseline refresh after verified sync
• Product ideation documentation — Updated README, changelog, and website docs to reflect the new BA orchestration flow, optional exploration, expanded artifact set, and verified Jira sync process

2026-05-17​

Changed​

• Cost optimization — Switched default model from Claude Opus 4.6 to GPT-5.4 across implementation and infrastructure agents (tsh-architect, tsh-engineering-manager) and public prompts (/tsh-implement, /tsh-analyze-aws-costs, /tsh-analyze-gcp-costs, /tsh-audit-infrastructure, /tsh-review-codebase)
• tsh-e2e-engineer agent — Changed model from Claude Sonnet 4.6 to GPT-5.4 mini
• Internal prompts — Removed YAML frontmatter (agent, model, description) from all internal prompts (tsh-deploy-kubernetes, tsh-implement-common-task, tsh-implement-e2e, tsh-implement-observability, tsh-implement-pipeline, tsh-implement-terraform, tsh-implement-ui-common-task, tsh-implement-ui, tsh-plan, tsh-research); internal prompts now fully inherit context from the delegating agent

2026-05-15​

Changed​

• Copilot customization agents — Updated model assignments: tsh-copilot-engineer and tsh-copilot-orchestrator switched from Claude Opus 4.6 to GPT-5.4; added explicit model to tsh-copilot-artifact-creator (GPT-5.4 mini), tsh-copilot-artifact-reviewer (Gemini 3.1 Pro), and tsh-copilot-researcher (Claude Sonnet 4.6)
• Copilot customization prompts (/tsh-create-custom-agent, /tsh-create-custom-instructions, /tsh-create-custom-prompt, /tsh-create-custom-skill) — Removed model field from frontmatter; prompts now inherit the model from the routed agent (tsh-copilot-orchestrator) instead of overriding it

2026-04-10​

Changed​

• /tsh-implement prompt — Fixed chronic UI verification skipping: added mandatory UI task inventory at plan review (step 2), proactive dev server URL collection before implementation (step 3), elevated [REUSE] UI verification to a prominent task type with explicit delegation instructions (step 6), added mandatory UI Verification Gate before code review (step 8), and explicit code review delegation step (step 9); references tsh-implement-ui.prompt.md for full verification workflow instead of duplicating it
• Engineering Manager agent (tsh-engineering-manager) — Added "UI Verification Enforcement" subsection with 4-point checklist (inventory at plan review, early dev server URL collection, process in order, gate code review); strengthened tsh-ui-reviewer delegation with mandatory emphasis and "never skip" guardrail
• UI Reviewer agent (tsh-ui-reviewer) — Added "Tool-to-URL mapping" rule clarifying that all Figma data (URLs, node IDs, file keys) must go through figma tool and Playwright is only for dev server navigation

2026-04-01​

Changed​

• Renamed Figma MCP server key from figma-mcp-server to figma across all agents, prompts, skills, MCP configuration, and documentation — aligns with Figma's recommended server naming in their official docs

2026-03-30​

Added​

• Backend development skill tsh-implementing-backend

Changed​

• Updated tsh-implementing-backend skill reference in tsh-software-engineer agent
• Updated tsh-implementing-backend as a conditional skill in implement prompt for backend API tasks

2026-03-20​

Changed​

• /tsh-implement prompt — Now auto-detects missing context and missing plan; delegates to tsh-context-engineer for research and tsh-architect for planning before implementation, with user confirmation between phases
• /tsh-plan prompt — Moved from public .github/prompts/ to internal .github/internal-prompts/; no longer invoked directly by users — the Engineering Manager delegates to the Architect automatically when a plan is needed
• /tsh-research prompt — Moved from public .github/prompts/ to internal .github/internal-prompts/; no longer invoked directly by users — the Engineering Manager delegates to the Context Engineer automatically when research is needed
• Engineering Manager agent (tsh-engineering-manager) — Added tsh-context-engineer to subagents; added structured workflow to decide between research, planning, and implementation phases; added delegation rules for tsh-context-engineer (missing context) and tsh-architect (missing plan); added Sequential Thinking usage for phase routing decisions
• Business Analyst agent (tsh-business-analyst) — Replaced "Deep-dive Research per Task" and "Prepare Implementation Plan" handoff buttons with single "Start Implementation" handoff routing to Engineering Manager
• Context Engineer agent (tsh-context-engineer) — Replaced "Prepare Implementation Plan" handoff button with "Start Implementation" handoff routing to Engineering Manager
• Updated website documentation: moved /tsh-plan and /tsh-research prompt pages from public to internal section; updated agents overview, prompts overview, workflow docs, and getting started pages

2026-03-17​

Added​

• Engineering Manager agent (tsh-engineering-manager) — Orchestrates the implementation phase by delegating tasks to specialized agents (Software Engineer, E2E Engineer, DevOps Engineer, Architect, Code Reviewer, UI Reviewer) based on the implementation plan; uses Sequential Thinking for ambiguous routing; auto-triggers code review if no review phase is defined; tracks progress via plan checkboxes
• Internal prompts directory (.github/internal-prompts/) — Agent-only prompts not visible in the slash command menu, used exclusively for sub-agent delegation by the Engineering Manager
• Internal prompt tsh-implement-common-task — Base implementation workflow for Software Engineer delegated tasks (backend and non-Figma frontend)
• Internal prompt tsh-implement-ui-common-task — Extends tsh-implement-common-task with UI-specific behaviors for Figma-based frontend tasks
• Internal prompt tsh-implement-ui — Full UI implementation + verification loop orchestration for the Engineering Manager
• Documentation page for the Engineering Manager agent on the website
• Documentation pages for all new internal prompts on the website

Changed​

• /tsh-implement prompt — Rewritten to route through the Engineering Manager agent instead of Software Engineer; now delegates tasks to specialized agents based on plan task types ([CREATE], [MODIFY], [REUSE])
• Architect agent (tsh-architect) — Handoff now routes to Engineering Manager instead of Software Engineer; removed "Start UI Implementation" handoff button (consolidated into single "Start Implementation"); reformatted tools list YAML; updated plan template to include [REUSE] UI verification tasks delegated to tsh-ui-reviewer
• Architecture Designing skill (tsh-architecture-designing) — Updated plan phases to run only fast tests/checks per phase (unit, integration, linters, build); added code review phase requirement using tsh-code-reviewer with tsh-review.prompt.md; added [REUSE] UI verification task pattern for Figma-based features
• UI Reviewer agent (tsh-ui-reviewer) — Removed "Start UI Implementation" and "Implement UI Fixes" handoff buttons (Engineering Manager now owns the verify-fix loop); added explicit dev server URL confirmation requirement; added authentication/login screen detection and escalation; added "reading source code is NOT verification" guardrail
• Code Reviewer agent (tsh-code-reviewer) — Added explicit mention of e2e tests alongside unit and integration tests in verification requirements
• Software Engineer agent (tsh-software-engineer) — Removed atlassian/search from tool access (Atlassian context now gathered by Engineering Manager)
• /tsh-plan prompt — Minor update
• /tsh-review-ui prompt — Minor update
• /tsh-review prompt — Minor update
• Prompts reorganized into public and internal categories on the documentation website with separate sidebar sections
• Moved 7 infrastructure/DevOps prompts from public .github/prompts/ to internal .github/internal-prompts/ (tsh-deploy-kubernetes, tsh-implement-e2e, tsh-implement-observability, tsh-implement-pipeline, tsh-implement-terraform)
• Updated agents overview documentation with Engineering Manager in the handoff diagram and agent summary table
• Updated prompts overview documentation with public/internal prompt distinction and delegation table
• Updated workflow documentation (standard flow, frontend flow, e2e flow) to reflect Engineering Manager orchestration

Removed​

• /tsh-implement-ui public prompt — Consolidated into /tsh-implement; UI implementation is now handled internally by the Engineering Manager's delegation to Software Engineer + UI Reviewer
• /tsh-clean-transcript prompt — Removed (functionality available through /tsh-analyze-materials)
• /tsh-create-jira-tasks prompt — Removed (functionality available through /tsh-analyze-materials)

2026-03-08​

Added​

• Ensuring Accessibility skill (tsh-ensuring-accessibility) — WCAG 2.1 AA compliance, semantic HTML, ARIA patterns, keyboard navigation, focus management, screen reader support, and color contrast requirements
• Implementing Forms skill (tsh-implementing-forms) — Form architecture, schema-based validation, field composition, error handling, multi-step form flows, and accessible form patterns
• Frontend Optimization skill (tsh-optimizing-frontend) — Rendering optimization, code splitting, memoization strategies, bundle size control, asset optimization, and memory management with React-specific reference patterns
• Frontend Review skill (tsh-reviewing-frontend) — Frontend-specific code review criteria: component anti-patterns, hooks quality, rendering correctness, accessibility and performance spot-checks, module organization with React-specific reference checklist
• Writing Hooks skill (tsh-writing-hooks) — Custom hook and composable patterns: naming, composition, stable return shapes, lifecycle cleanup, and testing strategies with React-specific reference patterns
• React-specific reference files (references/react-patterns.md) for implementing-frontend, optimizing-frontend, reviewing-frontend, and writing-hooks skills
• Documentation pages for all 5 new skills on the website

Changed​

• Software Engineer agent (tsh-software-engineer) — Added 4 new frontend skills to skills list (tsh-implementing-forms, tsh-writing-hooks, tsh-ensuring-accessibility, tsh-optimizing-frontend); added tsh-ui-reviewer as subagent for verification delegation; reformatted tools list
• Code Reviewer agent (tsh-code-reviewer) — Added tsh-reviewing-frontend skill for frontend-specific review criteria
• UI Reviewer agent (tsh-ui-reviewer) — Rewritten to emphasize subagent usage pattern, mandatory tool-based verification (never mental comparison), transparent error reporting with LOW confidence; reformatted tools list
• Frontend Implementation skill (tsh-implementing-frontend) — Refactored to focus on component patterns and composition, moved accessibility to dedicated tsh-ensuring-accessibility skill; added React-specific reference file
• UI Verification skill (tsh-ui-verifying) — Rewritten with 5-step verification process, verification order (stop on first CRITICAL failure), and improved report format
• /tsh-implement-ui prompt — Rewritten to use tsh-ui-reviewer as subagent (not /tsh-review-ui prompt call); added tsh-ensuring-accessibility skill; clarified that SE must never verify UI itself
• /tsh-review-ui prompt — Simplified to delegate entirely to tsh-ui-verifying skill workflow; fixed "all differences" wording to align with skill's stop-on-critical-failure rule
• Updated website documentation for Software Engineer, Code Reviewer, UI Reviewer agents and /tsh-implement-ui, /tsh-review-ui prompts
• Updated skills overview: skill count 25 → 30, added new skills to Development and Quality tables, updated agent–skill matrix
• Fixed Architect agent docs — added 7 missing skills (multi-cloud, cloud cost, CI/CD, Terraform, secrets, Kubernetes, observability)
• Fixed DevOps Engineer agent docs — added missing tsh-codebase-analysing skill
• Fixed Frontend Flow workflow docs — added tsh-ensuring-accessibility to required skills, updated subagent terminology

2026-03-06​

Added​

• DevOps Engineer agent (tsh-devops-engineer) — Senior DevOps Engineer and Consultant persona specializing in Golden Paths, automation, and Cloud governance; mandatory architect sub-agent delegation for all design decisions; multi-cloud guardrails with FinOps alerts (>10% cost increase triggers alert); three-option output strategy (Golden Path, Cost-Optimized, Velocity); mandatory skill-loading chains for 8 task types; tools include AWS API MCP, AWS Docs MCP, GCP gcloud/observability/storage MCPs, Context7, Sequential Thinking
• Multi-Cloud Architecture skill (tsh-designing-multi-cloud-architecture) for selecting and integrating services across AWS, Azure, and GCP with service comparison and multi-cloud pattern references
• CI/CD Implementation skill (tsh-implementing-ci-cd) for pipeline design patterns and deployment strategies
• Kubernetes Implementation skill (tsh-implementing-kubernetes) for deployment patterns, Helm charts, and cluster management
• Observability Implementation skill (tsh-implementing-observability) for logging, monitoring, alerting, and distributed tracing patterns
• Terraform Modules skill (tsh-implementing-terraform-modules) for reusable Terraform modules across AWS, Azure, and GCP with per-cloud module references
• Secrets Management skill (tsh-managing-secrets) for secrets management patterns in cloud and Kubernetes environments
• Cloud Cost Optimization skill (tsh-optimizing-cloud-cost) for rightsizing, tagging strategies, and spending analysis with tagging standards reference
• AWS cost analysis prompt (/tsh-analyze-aws-costs) for cost optimization and tagging compliance audit with hybrid IaC + live API approach
• GCP cost analysis prompt (/tsh-analyze-gcp-costs) for cost optimization and labeling compliance audit with hybrid IaC + live API approach
• Infrastructure audit prompt (/tsh-audit-infrastructure) for multi-scope audit (AWS/Azure/GCP/K8s/CI-CD) covering security, cost, and best practices
• Kubernetes deployment prompt (/tsh-deploy-kubernetes) for deployments, Helm charts, and workload configurations
• CI/CD pipeline prompt (/tsh-implement-pipeline) for pipelines with deployment stages and environment protection
• Terraform implementation prompt (/tsh-implement-terraform) for Terraform modules and cloud infrastructure provisioning
• Observability implementation prompt (/tsh-implement-observability) for metrics, logs, traces, and alerting solutions

Changed​

• Updated Architect agent (tsh-architect) with handoff to DevOps Engineer for infrastructure implementation
• Renamed 7 new infrastructure skill directories with tsh- prefix (continuation of 2026-03-05 prefix migration)
• Renamed 7 new infrastructure prompt files with tsh- prefix
• Updated all skill cross-references in architect agent, devops engineer agent, and all 7 infrastructure SKILL.md files
• Updated all skill references in 7 infrastructure prompt files

2026-03-05​

Changed​

• Added tsh- prefix to all Copilot customization artifacts to prevent naming collisions when used alongside project-specific customizations
• Renamed all 18 skill directories to include tsh- prefix (e.g., code-reviewing → tsh-code-reviewing, creating-agents → tsh-creating-agents)
• Renamed all 15 prompt files to include tsh- prefix (e.g., /create-custom-agent → /tsh-create-custom-agent, /implement → /tsh-implement)
• Renamed worker agents to include tsh- prefix: copilot-researcher → tsh-copilot-researcher, copilot-artifact-creator → tsh-copilot-artifact-creator, copilot-artifact-reviewer → tsh-copilot-artifact-reviewer
• Updated all cross-references between artifacts to use prefixed names

Added​

• Naming convention instruction (.github/instructions/naming-conventions.instructions.md) enforcing tsh- prefix on all artifact filenames, frontmatter names, and cross-references
• tsh- prefix explanation note in README for external users

2026-03-02​

Added​

• Custom agent creation prompt (/create-custom-agent) for creating new .agent.md files via the orchestrator — researches existing patterns, guides design decisions, creates and validates the agent file
• Custom skill creation prompt (/create-custom-skill) for creating new SKILL.md files via the orchestrator — enforces gerund naming, creates supporting resources alongside the skill file
• Custom prompt creation prompt (/create-custom-prompt) for creating new .prompt.md files via the orchestrator — identifies correct agent routing, ensures prompt follows established patterns
• Custom instructions creation prompt (/create-custom-instructions) for creating new .instructions.md or copilot-instructions.md files via the orchestrator — helps decide between repo-level and file-scoped instructions

Changed​

• Creating Agents, Creating Skills, Creating Prompts, and Creating Instructions skills marked as internal (agent-only) — hidden from the slash command menu via user-invokable: false in SKILL.md frontmatter while remaining accessible to agents
• New /create-custom-* prompts serve as the recommended user-facing entry points for Copilot customization workflows, replacing direct skill invocation

2026-02-27​

Added​

• Copilot Engineer agent (tsh-copilot-engineer) for designing, creating, reviewing, and improving all GitHub Copilot customization artifacts — custom agents, skills, prompts, and instructions
• Copilot Orchestrator agent (tsh-copilot-orchestrator) for coordinating complex, multi-step Copilot engineering tasks by decomposing work into focused subtasks and delegating to specialized workers
• Copilot Researcher worker agent (copilot-researcher) for gathering, analyzing, and summarizing information from codebases and documentation — read-only research specialist for orchestrator delegation
• Copilot Artifact Creator worker agent (copilot-artifact-creator) for building and modifying Copilot customization artifacts based on detailed specifications — creation specialist for orchestrator delegation
• Copilot Artifact Reviewer worker agent (copilot-artifact-reviewer) for evaluating Copilot customization artifacts against best practices, workspace consistency, and structural correctness — review specialist for orchestrator delegation
• Orchestrator pattern documentation (docs/orchestrator-pattern.md) describing the orchestrator + specialized workers architecture as an alternative to monolithic agents, addressing context window degradation in complex multi-step tasks
• Creating Agents skill (creating-agents) with agent file template, structural conventions, and validation checklist for building .agent.md files
• Creating Skills skill (creating-skills) with naming conventions, body structure guidelines, progressive disclosure patterns, templates, and examples for building SKILL.md files
• Creating Prompts skill (creating-prompts) with prompt file template, workflow focus guidelines, and validation checklist for building .prompt.md files
• Creating Instructions skill (creating-instructions) with templates for repository-level and granular instruction files, decision framework for instruction vs. skill placement

Changed​

• Adopted gerund-form naming convention (verb-ing + object) as the standard for all skill directories, documented in README and enforced by the Creating Skills skill
• Existing skills will be adapted to follow the new gerund-form naming convention in separate upcoming pull requests

2026-02-24​

Added​

• Workshop Analyst agent (tsh-workshop-analyst) for converting discovery workshop materials (transcripts, designs, codebase context) into Jira-ready epics and user stories
• Transcript Processing skill (transcript-processing) for cleaning raw workshop/meeting transcripts and extracting structured business-relevant content
• Task Extraction skill (task-extraction) for identifying and structuring epics and user stories from workshop materials
• Task Quality Review skill (task-quality-review) for analyzing extracted tasks for quality gaps, missing edge cases, and improvement opportunities
• Jira Task Formatting skill (jira-task-formatting) for transforming extracted tasks into Jira-ready format with field mapping and markdown compatibility
• Workshop analysis prompts: /workshop-analyze, /transcript-clean, /create-jira-tasks

2026-02-18​

Added​

• SQL & Database engineering skill covering schema design (naming conventions, primary key strategies, data types, normalisation), performant SQL writing, indexing strategies, join optimisation, locking mechanics, transactions, query debugging with EXPLAIN ANALYZE, and ORM integration (TypeORM, Prisma, Doctrine, Eloquent, Entity Framework, Hibernate, GORM). Applies to PostgreSQL, MySQL, MariaDB, SQL Server, and Oracle

2026-02-17​

Added​

• Frontend Implementation skill (frontend-implementation) for accessibility, design system usage, component patterns, and performance guidelines
• UI Verification skill (ui-verification) for verification criteria, tolerances, checklists, and severity definitions

Changed​

• Consolidated tsh-frontend-software-engineer agent into tsh-software-engineer - frontend capabilities are now handled via skills
• Updated tsh-software-engineer tool guidelines with frontend-specific scenarios (Figma, Playwright, design tokens)
• Made skills tool-agnostic by removing hardcoded tool names
• Refactored implement-ui.prompt.md and review-ui.prompt.md to reference skills instead of duplicating content

Removed​

• tsh-frontend-software-engineer agent (replaced by tsh-software-engineer + frontend skills)

2026-02-15​

Added​

• Code quality check prompt (/code-quality-check) for comprehensive repository analysis covering dead code detection, duplication identification, improvement opportunities, and architecture review

2026-02-08​

Added​

• Technical context discovery skill for codebase exploration and understanding

Changed​

• Refactored agents, prompts, and skills to follow a consistent standard
• Improved architecture-design plan example with expanded detail
• Updated implementation-gap-analysis and task-analysis examples
• Streamlined agent definitions by extracting workflow logic into prompts and skills

2026-02-07​

Added​

• Skills support for modular, domain-specific agent capabilities (architecture-design, code-review, codebase-analysis, e2e-testing, implementation-gap-analysis, task-analysis)

Changed​

• Cleaned up repository structure

2026-02-05​

Changed​

• Switched default model to Claude Opus 4.6
• Updated documentation for VS Code 1.109 compatibility

2026-02-03​

Removed​

• GitHub MCP integration
• Copilot Spaces usage from agents

2026-01-29​

Fixed​

• Updated tool names to follow new VS Code naming pattern

2026-01-21​

Fixed​

• Updated Atlassian MCP URL to new recommended endpoint

2026-01-15​

Changed​

• Removed "(Preview)" label from model names in all prompt files for consistency

2026-01-08​

Changed​

• Updated package name

2026-01-07​

Changed​

• Updated agent tools for improved functionality and testing capabilities

2025-12-18​

Added​

• Frontend Software Engineer agent with UI implementation workflow
• UI implementation prompt with iterative UI verification process

2025-12-16​

Changed​

• Standardized tool names across all agents

2025-12-15​

Changed​

• Separated workflow instructions from agent identity definitions

2025-12-12​

Added​

• Language consistency guidelines for agents

Changed​

• Code reviewer now runs automatically after implementation

2025-12-11​

Added​

• Copilot Pro license requirement documentation

2025-12-10​

Changed​

• Updated review prompt model to Claude Opus 4.5

2025-12-08​

Added​

• Domain-specific Copilot Spaces support for agents
• Code reviewer as a subagent of the software engineer

2025-12-02​

Added​

• VS Code version requirement documentation (1.99+)

Changed​

• Generalized software engineer agent (previously backend-specific)
• Standardized agent descriptions and enforced instructions usage
• Switched agents to use Claude Opus

2025-11-28​

Added​

• Figma MCP Server integration for UI verification
• Git-committer agent with automated commit message generation

2025-11-26​

Added​

• tsh- prefix for all agent names for namespace consistency
• Atlassian resource accessibility checks

2025-11-23​

Added​

• Detailed MCP tool usage guidelines for all agents (Context7, Playwright, Figma, Atlassian)

2025-11-21​

Added​

• Sequential Thinking MCP integration for complex problem-solving
• Data Engineer agent

2025-11-20​

Added​

• MCP server configurations (Playwright, Context7, Figma Dev Mode, Atlassian)
• UI/Figma verification agent and review workflow
• Frontend Software Engineer agent (initial base)

2025-11-19​

Added​

• MCP configuration for workspace and user-level setups
• LICENSE file and updated README

2025-11-14​

Added​

• Agent-based architecture with handoffs (Architect, Business Analyst, Software Engineer, Code Reviewer)

Changed​

• Updated models to GPT-5.1 across prompts
• Specified Figma MCP usage in research workflow

2025-11-05​

Changed​

• Planning prompt now focuses on tasks only, excluding improvements

2025-11-03​

Added​

• New operational mode and additional tools

2025-10-31​

Changed​

• Narrowed Atlassian/Jira access scope
• Enhanced planning and research prompts with implementation analysis guidelines

2025-10-29​

Added​

• Plan prompt with task-specific implementation focus

2025-10-28​

Added​

• Initial project setup with EditorConfig, Prettier, Husky, and Copilot configurations
• Automated commit message generation prompt
• Security review configuration and documentation