Execute Hard Lock

Execute Hard Lock

Thin wrapper that resolves the hard-lock prompt via the drmCopilotExtension MCP tool and hands the resolved text to the atomic-executor subagent as kickoff directives. The resolved prompt is the authoritative instruction set for the session; this skill does not duplicate its contents.

When to Use This Skill

Use this skill when:

• The caller provides an explicit plan file path (${plan-path}) and a selected work mode (${work-mode}).
• Strict plan-following behavior is required (no replanning, no reordering, no bucket tasks).
• The drmCopilotExtension MCP server is registered and reachable.

Inputs

Required:

• plan-path — absolute or repo-relative path to the plan-of-record markdown file.
• work-mode — one of minor-audit, full-feature, full-bug. Legacy value full normalizes to full-feature. Missing or malformed values fail closed to full-feature.

Invocation Flow

1. Resolve the Hard-Lock Prompt via MCP

Call the extension's resolver as the first action:

• Tool: mcp__drmCopilotExtension__resolve_execute_hard_lock_prompt
• Parameters:
  • target (required): the plan-of-record path (${plan-path}).
  • workspace_root (optional): the workspace root. Omit to default to the current working directory.

On success, the response contains an artifacts array whose first entry is the absolute path of a file containing the resolved hard-lock prompt (produced by the extension passing --output and --quiet to the bundled Python resolver).

2. Read the Resolved Prompt

Use the Read tool on the path returned in artifacts[0]. Treat the file contents as the authoritative hard-lock instruction set for this session.

3. Delegate to atomic-executor

Invoke the atomic-executor subagent via the Agent tool. Pass the resolved prompt text as the subagent's kickoff directive, followed by ${plan-path} and ${work-mode} as session context.

The resolved prompt itself already instructs the subagent to perform the mandatory read-proof (git rev-parse HEAD, SHA-256 of the plan file, unchecked-task count), preflight validation, and READY TO BEGIN FROM [P#-T#] handshake before executing any task. This skill does not reissue those instructions locally — doing so would risk divergence from the canonical template.

Abort Conditions

Stop immediately and report BLOCKED: execute-hard-lock <cause> in any of these cases. Do not reconstruct the hard-lock prompt from any other source and do not delegate to atomic-executor without a successful read in step 2.

• MCP tool is not available or not permitted.
• MCP response has ok: false or is otherwise malformed.
• MCP response omits artifacts or artifacts[0].
• Read on the artifact path fails (file missing, unreadable, empty).

Equivalent Entry Points (Reference)

The three entry points below all produce the same resolved hard-lock prompt for a given plan path. This skill always uses the MCP form:

• MCP (used by this skill): mcp__drmCopilotExtension__resolve_execute_hard_lock_prompt with target=<plan-path>. The extension passes --output artifacts/hard_lock_prompt.txt and --quiet to the bundled Python resolver.
• VS Code command: @command:drmCopilotExtension.resolveExecuteHardLockPrompt (interactive; writes to stdout + clipboard, no file artifact).

Delegation Contract

The atomic-executor subagent at .claude/agents/atomic-executor.md enforces the persistent execution-mode behaviors (anti-replanning, preflight-only blocking, persistence across turns, resume) and preloads the shared skills that support them:

• policy-compliance-order — mandatory policy reading order.
• atomic-plan-contract — plan format, Phase 0 requirements, preflight signals, validator gate, and mode-specific plan gates.
• acceptance-criteria-tracking — AC check-off protocol and status summary format.
• evidence-and-timestamp-conventions — baseline and final-QC artifact paths and required fields.

Prohibitions

• Do not proceed without a successful MCP resolver response AND a successful Read of the artifact.
• Do not reconstruct the hard-lock contract from any other source (not from this file, not from .github/prompts/execute-hard-lock.prompt.md, not from prior session memory).
• Do not modify the resolved prompt text before passing it to atomic-executor.
• Do not replan, reorder, or add tasks at this layer — the subagent owns that contract.
• Do not modify policy files under .claude/rules/ or .github/instructions/.
• Do not create or read secrets unless explicitly authorized.