SDD Review
This portable skill performs an independent code review of one task in review_required: an
explicit base and target diff scope against the approved contracts (requirements, stories,
architecture), the task evidence, and the project rules. Read templates/codereview.md and render it, keeping its
frontmatter keys and headings exactly (translate prose only).
Language: before writing human-facing prose, run scripts/sdd_language.py <repo-root> and use the persisted language; on not_initialized, return it with next_action: run_init. Localization rules: references/language.md.
Procedure
- Validate that the declared base, target, and changed paths are explicit and confined. Stop on
an ambiguous or expanded diff scope.
- Read only the approved contracts, current task evidence, and applicable project rules. Do not
change them or infer approval from their existence.
- Run and record the exact verification commands, environment, exit codes, and confined evidence
paths.
- Check skill conformity: confirm the change was produced through the SDD task contract and its
skills, not around them (no adapter bypass, no edited approvals).
- Record every finding with a stable ID, severity (
BLOCKER, HIGH, MEDIUM, LOW, INFO),
file and line, evidence, and rule citations or contract citations. Do not silently fix, hide,
or rewrite a finding; an unapproved change requires a new review.
- Produce an OKF v0.2
CODE_REVIEW report. A complete human-approved report records
scripts/sdd_tasks.py evidence <state> <TASK-ID> review --status approved and transitions to
verify_required; any blocker, missing evidence, scope ambiguity, contract violation, or
non-conformity records review --status rejected and transitions to rejected --reason <text>.
Read when
references/quality.md — writing the report (evidence and gate semantics shared with QA).
Output
Return the report path, exact base/target scope, approved contracts consumed, findings and
severities, cited rules, commands and exit codes, skill conformity, blocker status, the
no-silent-fix result, lifecycle status, and permitted next transition. Keep generation and
verification actors separate. Do not edit source, requirements, stories, architecture, task
evidence, or the review to make checks pass, and do not stop user-owned services.
Safety: Do not commit, push, or publish. Do not read or expose .env, credentials, tokens, private keys, certificates, or fleet environment files. Full contract: references/safety.md.
1---2name: sdd-review3description: >4 Independent, traceable code review of one SDD Composy task in review_required5 against its approved PRD, stories, TechSpec, evidence, and project rules, with6 stable findings and severities. Use when a task passed QA and evidence — 'revisar7 a TASK-003', 'code review da task', 'review this task's diff'. Do NOT use for code8 review outside an SDD workspace (use the host code review), for QA (sdd-qa), or9 for verification (sdd-verify).10metadata:11 source: https://skillmd.com/skills/pwdev-solucoes/sdd-review12 version: 0.1.013---1415# SDD Review1617This portable skill performs an independent code review of one task in `review_required`: an18explicit base and target diff scope against the approved contracts (requirements, stories,19architecture), the task evidence, and the project rules. Read `templates/codereview.md` and render it, keeping its20frontmatter keys and headings exactly (translate prose only).2122Language: before writing human-facing prose, run `scripts/sdd_language.py <repo-root>` and use the persisted language; on `not_initialized`, return it with `next_action: run_init`. Localization rules: `references/language.md`.2324## Procedure25261. Validate that the declared base, target, and changed paths are explicit and confined. Stop on27 an ambiguous or expanded diff scope.282. Read only the approved contracts, current task evidence, and applicable project rules. Do not29 change them or infer approval from their existence.303. Run and record the exact verification commands, environment, exit codes, and confined evidence31 paths.324. Check skill conformity: confirm the change was produced through the SDD task contract and its33 skills, not around them (no adapter bypass, no edited approvals).345. Record every finding with a stable ID, severity (`BLOCKER`, `HIGH`, `MEDIUM`, `LOW`, `INFO`),35 file and line, evidence, and rule citations or contract citations. Do not silently fix, hide,36 or rewrite a finding; an unapproved change requires a new review.376. Produce an OKF v0.2 `CODE_REVIEW` report. A complete human-approved report records38 `scripts/sdd_tasks.py evidence <state> <TASK-ID> review --status approved` and transitions to39 `verify_required`; any blocker, missing evidence, scope ambiguity, contract violation, or40 non-conformity records `review --status rejected` and transitions to `rejected --reason <text>`.4142## Read when4344- `references/quality.md` — writing the report (evidence and gate semantics shared with QA).4546## Output4748Return the report path, exact base/target scope, approved contracts consumed, findings and49severities, cited rules, commands and exit codes, skill conformity, blocker status, the50no-silent-fix result, lifecycle status, and permitted next transition. Keep generation and51verification actors separate. Do not edit source, requirements, stories, architecture, task52evidence, or the review to make checks pass, and do not stop user-owned services.5354Safety: Do not commit, push, or publish. Do not read or expose `.env`, credentials, tokens, private keys, certificates, or fleet environment files. Full contract: `references/safety.md`.