/adversarial-review

Run an adversarial OpenCode review through the shared built-in reviewer.

Raw slash-command arguments: $ARGUMENTS

Core constraint:

• This command is review-only.
• Do not fix issues, apply patches, or suggest that you are about to make changes.
• Your only job is to run the review and return OpenCode's output verbatim to the user.

Execution mode rules:

• If the raw arguments include --wait, do not ask. Run the review in the foreground.
• If the raw arguments include --background, do not ask. Run the review in a Claude background task.
• Otherwise, estimate the review size before asking:
  • For working-tree review, start with git status --short --untracked-files=all.
  • For working-tree review, also inspect both git diff --shortstat --cached and git diff --shortstat.
  • For base-branch review, use git diff --shortstat <base>...HEAD.
  • For PR review (--pr <N> or a PR #N reference in the focus text), use gh pr view <N> --json additions,deletions,changedFiles and base the recommendation on those numbers (large PRs almost always go to background).
  • Treat untracked files or directories as reviewable work even when git diff --shortstat is empty.
  • Only conclude there is nothing to review when the relevant working-tree status is empty or the explicit branch diff is empty.
  • Recommend waiting only when the review is clearly tiny, roughly 1-2 files total and no sign of a broader directory-sized change.
  • In every other case, including unclear size, recommend background.
  • When in doubt, run the review instead of declaring that there is nothing to review.
• Then use AskUserQuestion exactly once with two options, putting the recommended option first and suffixing its label with (Recommended):
  • Wait for results
  • Run in background

Argument handling:

• Preserve the user's arguments exactly.
• Do not strip --wait, --background, --model, --free, or --pr yourself.
• Adversarial reviews support custom focus text. Any text after flags is treated as a focus area.
• The companion script handles --adversarial internally.
• --model <id> overrides OpenCode's default model for this single review (e.g. --model openrouter/anthropic/claude-opus-4-6). Pass it through verbatim if the user supplied it.
• --free tells the companion script to shell out to opencode models, filter for first-party opencode/* free-tier models (those ending in :free or -free), and pick one at random for this review. Restricted to the opencode/* provider because OpenRouter free-tier models have inconsistent tool-use support, and the review agent needs read/grep/glob/list. Pass it through verbatim if the user supplied it. --free and --model are mutually exclusive — the companion will error if both are given.
• --pr <number> reviews a GitHub pull request via gh pr diff instead of the local working tree. The cwd must be a git repo whose remote points at the PR's repository, and gh must be installed and authenticated.

PR reference extraction (REQUIRED — read this carefully):

• If the user's input contains a PR reference like PR #390, pr #390, PR 390, or pr 390, you MUST extract the number yourself and pass it as --pr 390. Then strip the matched PR phrase from whatever you put in the focus text.
• Do not rely on the companion script's focus-text auto-detection for PR refs. Bash strips unquoted #NNN tokens as comments before they ever reach the companion script (e.g. bash -c 'node ... adversarial-review on PR #390' invokes node with argv ["adversarial-review", "on", "PR"] — the #390 is gone).
• Example transformation:
  • User input: /opencode:adversarial-review on PR #390 challenge the caching strategy
  • Wrong: node ... adversarial-review on PR #390 challenge the caching strategy (the #390 and everything after is silently dropped by bash)
  • Right: node ... adversarial-review --pr 390 'challenge the caching strategy'

Focus text quoting (REQUIRED):

• When you invoke the companion script via Bash and the user's focus text contains shell metacharacters (#, *, $, ;, &, |, <, >, parentheses, backticks, etc.), wrap the focus text in single quotes so bash passes it through unchanged. Single quotes preserve everything literally except the single-quote character itself.
• If the focus text itself contains a single quote (e.g. what's wrong here), use the standard '\'' escape: 'what'\''s wrong here'.
• Examples:
  • /opencode:adversarial-review challenge the design → node ... adversarial-review 'challenge the design'
  • /opencode:adversarial-review --background look for race conditions in $RUNTIME → node ... adversarial-review --background 'look for race conditions in $RUNTIME'

Foreground flow:

• First, transform $ARGUMENTS using the PR reference extraction and Focus text quoting rules above. Pass through --wait, --background, --base, --scope, --model, and --pr flags as-is; convert any PR #N reference in the user's text to --pr N; single-quote whatever free-form focus text remains.
• Then run the resulting command (illustrative shape — substitute the actual transformed args):

node "${CLAUDE_PLUGIN_ROOT}/scripts/opencode-companion.mjs" adversarial-review <flags> [--pr N] ['<quoted focus text>']

• Return the command stdout verbatim, exactly as-is.
• Do not paraphrase, summarize, or add commentary before or after it.
• Do not fix any issues mentioned in the review output.

Background flow:

• Apply the same $ARGUMENTS transformation as the foreground flow above (PR ref extraction + focus text single-quoting).
• Then launch the resulting command with Bash in the background:

Bash({
  command: `node "${CLAUDE_PLUGIN_ROOT}/scripts/opencode-companion.mjs" adversarial-review <flags> [--pr N] ['<quoted focus text>']`,
  description: "OpenCode adversarial review",
  run_in_background: true
})

• Do not call BashOutput or wait for completion in this turn.
• After launching the command, tell the user: "OpenCode adversarial review started in the background. Check /opencode:status for progress."