code-report

# code-report > **Stage-session entry (`standard+` dispatch, spec/stage-dispatch SD-2)**: Run in-session or as an isolated dispatch-depth-2 stage worker dispatched by the `autopilot-code` conductor. Read the plan, checklist, `dev_logs/`, `test_logs/`, `_internal/*_reviews/`, and `pipeline_summary.md` from disk; never assume prior-stage conversation is available. The write class is `final_report.md`, `<artifact-root>/analysis_project/code/*.md`, and lock-protected `pipeline_summary.md`. The stage runs as the `editorial/report` unit (the node's unit); it dispatches no units itself. > **Plan resolution**: Treat [arguments-and-decisions.md#plan-resolution](../autopilot-code/references/arguments-and-decisions.md) as the single authority for resolving `$ARG`. > **Language rule**: Follow the audience and artifact language contract in [arguments-and-decisions.md#language-rule](../autopilot-code/references/arguments-and-decisions.md). The report template must be localized to the selected artifact language rather than fixed to the language of this skill file. ## Writer and QA Policy Use one portable `fast writer` at every rigor tier. Prior stages already review the plan, implementation, and tests; this stage synthesizes their artifacts. The incoming intensity-derived rigor or plan-frontmatter `qa_level` is context only. It does not change the writer role, add parallel writers, or open a report-review loop. | Rigor | Report action | |---|---| | Light | One fast writer | | Standard...