build-execution-map

build-execution-map

Analyze a set of GitHub issues for dependency relationships using AI-driven pairwise assessment, then produce a structured execution map JSON artifact that partitions issues into dependency-ordered dispatch groups.

When to Use

• Before parallel dispatch of multiple issues to prevent merge conflicts
• When sous-chef receives a parallel request for N ≥ 2 issues
• As a standalone pre-analysis step before running an implementation campaign

Arguments

Space-separated issue numbers (required, minimum 2), plus optional flags:

• --base-ref <branch> — base branch to compare against (default: main)
• --assess-review-approach — assess whether each issue would benefit from a review-approach research pass before implementation (default: inactive)
• --max-parallel <N> — maximum number of issues in any single parallel group (default: 6). Groups exceeding this cap are split into sequential sub-groups of at most N issues each.

Example: 101 103 102 --base-ref main --max-parallel 4 --assess-review-approach

Critical Constraints

NEVER:

• Modify any source code files
• Create files outside {{AUTOSKILLIT_TEMP}}/build-execution-map/ directory (relative to the current working directory)
• Assume issues are independent without analysis
• Launch implementation pipelines — this skill only produces the map
• Use the execution_map or execution_map_report token names with unspaced = (always use key = value format)
• Override the AI's parallelism judgment with mechanical rules
• Assume issues conflict based solely on file-name overlap without reading the issue descriptions
• Run subagents in the background (run_in_background: true is prohibited)
• Treat a medium-severity cross-assessment as grounds for deferral — only critical severity defers
• Emit has_deferred / deferred_count / dispatched_count with markdown decorators

ALWAYS:

• Use parallel subagents (up to 8) for issue fetching in Step 1
• Use model: "sonnet" for all subagents
• Write both JSON and markdown report outputs to {{AUTOSKILLIT_TEMP}}/build-execution-map/
• Emit execution_map and execution_map_report tokens with absolute paths (use $(pwd) to resolve the working directory prefix)
• Emit structured output tokens as the final lines of text output (plain text, no markdown decorators)
• Check the actual codebase when uncertain whether two issues' changes overlap
• Capture per-pair reasoning in pairwise_assessments for auditability
• Anchor all output paths to the current working directory