← ClaudeAtlas

sdd-plan-correctionslisted

SDD corrections when a code review must become traceable tasks; does not implement or alter the report.
dougcunha/TokenHound · ★ 3 · AI & Automation · score 79
Install: claude install-skill dougcunha/TokenHound
# Plan SDD corrections 1. Fix one `tasks/prd-[slug]/codereview_[num]/` by argument or context. Search only for missing values; multiple candidates require a choice. Require and read `codereview.md` once; preserve it. **Output:** exact, readable review, without mixing IDs between reports. 2. Classify every item as actionable, informational, or pending. In `APPROVED`, plan only what was requested; in `APPROVED WITH RESERVATIONS`, improvements require authorized scope; in `REJECTED`, cover violations, incompleteness, and failures. Unknown status allows only explicitly actionable findings. Preserve `CR-NN`; in legacy reports without IDs, assign a local ID by order and section. Inventory root and `done/` metadata to reuse tasks; finding identity is review path + ID. **Output:** every item assigned; no duplicate finding or invented decision. 3. Check evidence in the smallest necessary code/TechSpec section. Group only the same cause with a reviewable result. Model an acyclic DAG, limits, files, and tests; number new tasks after the largest number in root and `done/`. For desktop C#/.NET, omit E2E; preserve the obligation with relevant unit, integration, or manual acceptance. Missing environment/decision becomes an explicit pending item, not a discarded finding. **Output:** every actionable finding has a new/existing task, acceptance, and verification; no orphan task. 4. Read [references/TEMPLATE_TASK.md](references/TEMPLATE_TASK.md) in full when writing. Create revi