glm-adversarial-review

name	glm-adversarial-review
description	Use when asking GLM direct API for an adversarial review.
user-invocable	true

GLM Adversarial Review

Use the API reviewer GLM adversarial-review workflow. Current Codex builds expose it as api-reviewers:glm-adversarial-review in the skill picker; its skill frontmatter name is glm-adversarial-review; the command contract is plugins/api-reviewers/commands/glm-adversarial-review.md.

Run from the repository root:

node plugins/api-reviewers/scripts/api-reviewer.mjs approval-request --provider glm --mode adversarial-review --scope branch-diff --prompt "<focus>"
node plugins/api-reviewers/scripts/api-reviewer.mjs run --provider glm --mode adversarial-review --scope branch-diff --approval-token "<approval_token.value>" --lifecycle-events jsonl --prompt "<focus>"

If the user provides a base ref, add --scope-base REF before --prompt. <focus> is the user's review prompt or focus area. The review timeout default is 600000 ms; API_REVIEWERS_TIMEOUT_MS is the non-interactive fallback. Rendered prompts above the provider max_prompt_chars budget fail before launch with source_content_transmission: "not_sent"; use narrower scope shards or override API_REVIEWERS_MAX_PROMPT_CHARS only after confirming the provider accepts larger prompts. Direct API reviews send selected source content to an external provider. Before launching or retrying, render the approval-request output, request explicit approval with recommended_tool_justification, and pass approval_token.value to run with --approval-token only after approval. If approval is denied, follow denial_action and generate a relay prompt instead of running the external API command. Render the returned JobRecord, render external_review_launched as soon as it appears, then render external_review before the review result when present. If the JobRecord failed, report error_code, error_message, http_status when present, and suggested_action. Never print API-key values.