drm-copilot

extensions/drm-copilot provides repository automation, customization publishing, and MCP bridge for drm-copilot workflows. It exposes two workspace-facing adapter surfaces over the same bundled repo-automation workflows:

• VS Code commands for interactive editor use.
• A stdio MCP server for Codex and other MCP clients.

The bundled workflows continue to execute from extension package resources. They do not copy repo-local scripts into the destination workspace before execution.

VS Code Commands

The extension continues to contribute these stable command IDs:

• drmCopilotExtension.helloPython
• drmCopilotExtension.helloPowerShell
• drmCopilotExtension.collectCommitContext
• drmCopilotExtension.collectPrContext
• drmCopilotExtension.runCodexNativeConverter
• drmCopilotExtension.pushDownCopilotCustomizations
• drmCopilotExtension.pushDownCodexAndAgentsCustomizations
• drmCopilotExtension.newPotentialBugEntry
• drmCopilotExtension.newPotentialEntry
• drmCopilotExtension.linkParentChild
• drmCopilotExtension.potentialToIssue
• drmCopilotExtension.newActiveFeatureFolder
• drmCopilotExtension.resolvePolicyAuditTemplateAsset
• drmCopilotExtension.resolveExecuteHardLockPrompt
• drmCopilotExtension.resolveAtomicPlanPrompt
• drmCopilotExtension.syncAgentsFromInstructions
• drmCopilotExtension.listMcpTools
• drmCopilotExtension.newClaudeWorktreeSession
• drmCopilotExtension.removeSecondaryWorktrees

The interactive VS Code flows keep their current prompts and branch/file pickers, but now delegate through the shared repo-automation service used by the MCP bridge.

Sync AGENTS.md from Instructions

Use the drm-copilot: Sync AGENTS.md from Instructions command (command ID: drmCopilotExtension.syncAgentsFromInstructions) from the Command Palette to regenerate AGENTS.md in the active workspace. The bundled PowerShell script discovers all .github/instructions/*.instructions.md files under the active workspace root, aggregates their content in a deterministic order, and writes the consolidated result to AGENTS.md at the workspace root. This replaces any manual edits to AGENTS.md with a fully generated output derived from the workspace's canonical .github instruction files.

New Claude Worktree Session

Use the drm-copilot: New Claude Worktree Session command (command ID: drmCopilotExtension.newClaudeWorktreeSession) from the Command Palette to open an integrated PowerShell terminal that creates a new git worktree, navigates into it, optionally installs and activates the poetry environment when the workspace declares poetry, and then starts an interactive claude session. The command prompts for an optional objective to pass to claude as the initial prompt.

Injecting a pre-claude script

The command can run a local PowerShell script in the new worktree immediately before claude is started. This lets a repository carry its own pre-session setup (for example, copying machine-local configuration, seeding environment variables, or running a repo-specific bootstrap) in the local repository rather than centrally in the extension.

How it works:

• The script path is read from the drmCopilotExtension.newClaudeWorktreeSession.preClaudeScriptPath configuration setting.
• The default value is .claude/hooks/pre-claude-session.ps1, resolved relative to the new worktree root.
• After the worktree is created, navigated into, and (when applicable) the poetry environment is installed and activated, the command runs the configured script if it exists in the worktree, then starts claude.
• The script runs only when it exists: the command emits a runtime existence guard (if (Test-Path -LiteralPath '<path>') { & '<path>' }), so a missing script is not an error and the session proceeds directly to claude.

To use the default convention, commit a PowerShell script to the repository at .claude/hooks/pre-claude-session.ps1. Because the script lives in the repository, it is present in the new worktree after git worktree add and runs automatically on the next worktree session.

To use a different path, set the configuration value in the repository's .vscode/settings.json so the setting travels with the local repository:

{
  "drmCopilotExtension.newClaudeWorktreeSession.preClaudeScriptPath": ".claude/hooks/my-bootstrap.ps1"
}

Notes:

• The path is resolved relative to the worktree root and is embedded with PowerShell single-quote escaping, so paths containing spaces or apostrophes are preserved literally.
• Leaving the setting empty or whitespace-only suppresses the pre-claude step entirely; no additional command is sent.
• The script runs in the worktree's PowerShell session before claude takes over the terminal.

Remove Secondary Worktrees

Use the drm-copilot: Remove Secondary Worktrees command (command ID: drmCopilotExtension.removeSecondaryWorktrees) from the Command Palette to remove all secondary git worktrees of the current repository. The primary (main) worktree is never removed.

How it works:

• The command first shows a modal confirmation. It proceeds only when you select "Remove All"; any other choice (including dismissal) cancels the operation without issuing any git command.
• It enumerates worktrees via git worktree list --porcelain and excludes the primary worktree by position (the first reported block), so the primary is never selected for removal.
• Each secondary worktree is removed with NON-force git worktree remove <path>. The --force option is never used, so a worktree that cannot be fully removed (for example, one with modified or untracked files) fails the removal cleanly and is left intact rather than partially deleted.
• Locked worktrees and prunable worktrees (those whose working directory is missing on disk) are skipped with a reason and left intact; no removal is attempted for them.
• git worktree prune is not invoked automatically. Prunable worktrees are reported as skipped rather than pruned.
• A worktree that cannot be removed does not abort the operation; the command continues with the remaining worktrees.
• Removed and skipped outcomes (with reasons) are written to the drm-copilot output channel. A summary notification is shown when the command completes: an information message when everything was removed or no secondary worktrees were found, and a warning message when one or more worktrees were skipped.

MCP Server

The extension package also builds a stdio MCP server named drmCopilotExtension.

List MCP Tools

Use the drm-copilot: List MCP Tools command (command ID: drmCopilotExtension.listMcpTools) from the Command Palette to inspect the tools currently exposed by the drmCopilotExtension MCP server. The command shows the semantic tool names, each tool's description, and its required input fields in a Quick Pick list, and it also writes the full list to the drm-copilot output channel for reference.

Downstream Codex skills should depend on the MCP server name drmCopilotExtension, not on raw VS Code command IDs such as drmCopilotExtension.collectPrContext.

Exposed MCP Tools

• collect_commit_context
• collect_pr_context
• run_codex_native_converter
• push_down_copilot_customizations
• push_down_codex_and_agents_customizations
• new_potential_bug_entry
• new_potential_entry
• link_parent_child
• potential_to_issue
• new_active_feature_folder
• resolve_atomic_plan_prompt
• resolve_execute_hard_lock_prompt
• run_poshqc_format
• run_poshqc_analyze
• run_poshqc_test
• run_poshqc_analyze_autofix
• run_poshqc_suite
• resolve_policy_audit_template_asset
• validate_orchestration_artifacts

MCP Runtime Expectations

• MCP tools are fully non-interactive.
• workspace_root is required by all workspace-targeted tools; the MCP server cannot infer the calling agent's checkout, so callers must pass the absolute worktree/checkout root explicitly. An omitted value returns a structured ok: false error rather than silently defaulting.
• collect_pr_context requires an explicit base branch/ref in MCP mode.
• resolve_policy_audit_template_asset requires asset and optionally accepts target_path; valid selectors are template, code-review-template, feature-audit-template, and agents. When target_path is omitted, callers receive the bundled source path for the requested asset.
• Bundled scripts are resolved from extensions/drm-copilot/resources/... at runtime.
• Subprocesses are launched with explicit argv arrays and shell: false.

Codex Configuration Example

Build the extension package first:

npm --prefix extensions/drm-copilot run build

Then configure the repo checkout as an MCP server:

{
  "mcpServers": {
    "drmCopilotExtension": {
      "command": "node",
      "args": ["extensions/drm-copilot/out/mcp-server.js"]
    }
  }
}

If the server is launched from a different working directory, pass workspace_root explicitly in tool calls so the destination workspace stays deterministic.

MCP Input Summary

• collect_commit_context: required workspace_root
• collect_pr_context: required workspace_root, required base
• run_codex_native_converter: required workspace_root, required mode, required source_ecosystem, required source_root, optional selected_paths, optional destination_root, optional artifact_root, optional enable_repo_prompts
• push_down_copilot_customizations: required workspace_root
• push_down_codex_and_agents_customizations: required workspace_root, optional packs, optional csharp_variant, optional memory_mode
• new_potential_bug_entry: required workspace_root, required short_name
• new_potential_entry: required workspace_root, required short_name
• link_parent_child: required workspace_root, required child_issue_number, required parent_issue_number
• potential_to_issue: required workspace_root, required potential_path, promotion_type, work_mode
• new_active_feature_folder: required workspace_root, required feature_name, type, work_mode, optional issue_number
• resolve_policy_audit_template_asset: required workspace_root, required asset (template | code-review-template | feature-audit-template | agents), optional target_path
• resolve_atomic_plan_prompt: required workspace_root, required target
• resolve_execute_hard_lock_prompt: required workspace_root, required target
• run_poshqc_format: required workspace_root, optional scan_folders
• run_poshqc_analyze: required workspace_root, optional scan_folders
• run_poshqc_test: required workspace_root, optional scan_folders
• run_poshqc_analyze_autofix: required workspace_root, optional scan_folders
• run_poshqc_suite: required workspace_root, optional scan_folders
• validate_orchestration_artifacts: required workspace_root, required artifact_type, required artifact_path, optional require_complete

MCP Result Shape

MCP tool calls return structured JSON with:

• ok
• tool
• workspace_root
• artifacts when the workflow has deterministic or discovered output paths
• summary
• stderr_excerpt when a subprocess failure surfaces stderr diagnostics

Runtime Requirements

• All commands run in-process in TypeScript; no python interpreter is required for any extension or MCP command.
• PowerShell commands prefer pwsh and fall back to powershell on Windows when available.
• An open workspace folder is required for workspace-targeted VS Code commands.
• MCP clients must build the package so out/mcp-server.js exists before launching the server.

Resolve Execute Hard-Lock Prompt and Resolve Atomic Plan Prompt run entirely in-process in TypeScript; they no longer require a Python runtime.

Execution Model

The shared repo-automation service runs the formerly-Python commands in-process in TypeScript (no interpreter is spawned). The remaining bundled wrapper resources it executes are PowerShell scripts and data assets:

• resources/templates/new-potential-entry.ps1
• resources/templates/link-parent-child.ps1
• resources/templates/policy_audit/policy-audit.yyyy-MM-ddTHH-mm.md
• resources/templates/policy_audit/code-review.yyyy-MM-ddTHH-mm.md
• resources/templates/policy_audit/feature-audit.yyyy-MM-ddTHH-mm.md
• resources/templates/policy_audit/AGENTS.md
• resources/templates/run-poshqc-format.ps1
• resources/templates/run-poshqc-analyze.ps1
• resources/templates/run-poshqc-test.ps1
• resources/templates/run-poshqc-analyze-autofix.ps1
• resources/templates/run-poshqc-suite.ps1
• resources/powershell/PoshQC/

In-process commands (commit context, PR context, Codex-native converter, push-down publishers, new potential bug entry, potential-to-issue, new active feature folder, resolve hard-lock prompt, resolve atomic-plan prompt, and the helloPython smoke test) execute directly in TypeScript against the workspace and the bundled data payloads under resources/; they spawn no Python.

The VS Code command adapters and the MCP server both call that same service layer. This preserves backward compatibility for the command IDs while providing a semantic MCP tool surface for downstream automation.

resolve_policy_audit_template_asset and drmCopilotExtension.resolvePolicyAuditTemplateAsset are additive surface adapters over the same bundled policy-audit assets. They support the selectors template, code-review-template, feature-audit-template, and agents. In MCP mode, callers receive the canonical asset id plus the bundled source path and, when requested, the copied destination path. In VS Code, interactive use opens the bundled asset when no target is supplied and copies it into the workspace when -target <path> is supplied.

Codex-native converter wrapper

Use the drm-copilot: Run Codex-native Converter command (command ID: drmCopilotExtension.runCodexNativeConverter) from the Command Palette to run the in-process TypeScript converter against the active workspace. The command prompts for review or apply mode and the source ecosystem, then runs the converter in-process.

The same workflow is available through the semantic MCP tool run_codex_native_converter. MCP callers provide the non-interactive input contract directly:

• required mode: review or apply
• required source_ecosystem: github-copilot or claude
• required source_root: workspace-relative or absolute source runtime root
• optional selected_paths: repeated workspace-relative or absolute source paths to limit the run scope
• optional destination_root: required when mode is apply
• optional artifact_root: override for the report-set root
• optional enable_repo_prompts: allow .codex/prompts/** output when the destination repository explicitly uses that surface

The command adapter remains thin by design: it does not implement a second converter. It passes the resolved input to the in-process TypeScript converter, surfaces the artifact root, and relies on the converter's fail-closed validation model before any destination writes occur.

Push Down Codex and Agents Customizations

Use the drm-copilot: Push Down Codex and Agents Customizations command (command ID: drmCopilotExtension.pushDownCodexAndAgentsCustomizations) from the Command Palette to copy the bundled .codex and .agents payload into the active workspace. The in-process TypeScript publisher uses the extension-packaged payload root at resources/codex-and-agents-customizations/ and writes a JSON summary artifact under artifacts/codex-and-agents-customizations/ in the destination workspace.

The command and MCP tool accept optional Codex language-pack selection. Omitted or empty packs publishes the complete .codex and .agents trees. When packs is supplied, core is included automatically. Supported public pack names are core, python, powershell, typescript, and csharp. Use csharp_variant to select the modern or legacy C# content.

The optional csharp_variant field accepts modern or legacy. The modern value uses the root Codex C# profile. The legacy value reads C# content from the bundle-only .agents-variants/csharp-legacy/ and .codex-variants/csharp-legacy/ roots and writes it to the canonical .agents and .codex destination paths. The variant roots are not destination roots.

The optional memory_mode field accepts overwrite, merge, or skip for schema parity with Claude push-down. It is inert for the current Codex bundle because no Codex memory subtree is present.

Run PoshQC Suite

Use the drm-copilot: Run PoshQC Suite command (command ID: drmCopilotExtension.runPoshQCSuite) from the Command Palette to run the bundled PowerShell quality gate from extension resources against the destination workspace. The bundled PowerShell wrapper executes resources/templates/run-poshqc-suite.ps1, imports the colocated resources/powershell/PoshQC/ module copy, and can limit scanning to one or more selected destination-workspace folders.

Additional granular bundled PoshQC surfaces are also available through both VS Code commands and semantic MCP tools:

• drmCopilotExtension.runPoshQCFormat / run_poshqc_format
• drmCopilotExtension.runPoshQCAnalyze / run_poshqc_analyze
• drmCopilotExtension.runPoshQCTest / run_poshqc_test
• drmCopilotExtension.runPoshQCAnalyzeAutofix / run_poshqc_analyze_autofix

Each command/tool uses the same workspace_root plus optional scan_folders contract as the bundled suite. The autofix operation is intentionally mutating: it applies PSScriptAnalyzer -Fix to the selected PowerShell files, then reruns bundled analysis and fails if findings remain.