Review

Review the complete task diff

OpenCode v1: Skill names below are exact IDs from the active catalog, not slash commands. Load them with the native skill tool. Slash commands are direct user entry points only.

This is the ordinary self-review pass. An independent Inspector verdict remains a separate review when the calling workflow requires it.

If the user requests a walkthrough, start explaining the change at their requested pace instead of replacing that request with this self-review pass. the full-context guided human review workflow is a manual skill: apply it when the user explicitly invokes it through the client's skill entrypoint. Do not automatically load it or require another confirmation or a skill command before providing a requested explanation.

Establish the boundary

Read the acceptance criteria, locked decisions, and repository instructions that govern the task. Resolve the stated base or fixed point, then account for staged, unstaged, committed, and untracked changes in scope. Read every new untracked file in full because ordinary diff output omits it.

Stop when the diff is empty, the base is ambiguous, or unrelated work cannot be separated safely. Review primary evidence, not the implementation summary or memory of the intended change.

Review six axes separately

1. Correctness — trace changed control and data flow; check errors, state transitions, cleanup, concurrency, and failure handling.
2. Regression — find affected callers, contracts, shared defaults, migrations, and behavior outside the immediate edit.
3. Edge cases — test relevant empty, missing, malformed, repeated, concurrent, timeout, retry, and partial-failure cases.
4. Requirement coverage — map every acceptance criterion and locked decision to concrete code or test evidence.
5. Repository standards — check established architecture, naming, language, test, documentation, and delivery rules.
6. Scope — require every hunk to serve the task and identify requested behavior the diff does not implement.

When the change touches personal or confidential data, authentication or authorization, secrets, logging, network trust, external integrations, deployment permissions, or another security boundary, invoke and apply security-review.

Correct and verify

If the active caller is the writer, fix in-scope findings before returning. When reviewing delegated work, do not edit the implementation in the review context; return the smallest bounded correction to the active writer instead.

Discover deterministic gates from repository instructions, build files, test configuration, and CI rather than assuming a toolchain. Run the smallest relevant checks and required final gates after the last edit. Record commands, relevant results, and exit codes; distinguish fresh evidence from stale or unverified claims.

Report material findings first, ordered by consequence. For each finding, give the affected path and line when available, the concrete failure mode, and the smallest useful correction. End with acceptance-criterion coverage, gate evidence, and remaining uncertainty. A clean review must still state what was inspected and what the evidence does not prove.