oh-my-opencode 5.0.0-beta.7 → 5.0.0-beta.8

package/.agents/skills/work-with-pr/SKILL.md

@@ -196,12 +196,20 @@ while true:
 
 ### Gate A: CI Checks
 
-CI is the fastest feedback loop. Wait for it to complete, then parse results.
+CI is the fastest feedback loop. Subscribe to its completion via `monitor` — never block a model round-trip on `gh pr checks --watch`.
 
-```bash
-# Wait for checks to start (GitHub needs a moment after push)
-# Then watch for completion
-gh pr checks "$PR_NUMBER" --watch --fail-fast
+```
+# Subscribe to CI completion — the monitor event wakes the session when checks finish.
+# Do NOT use `gh pr checks --watch` as a blocking tool call; it burns a full model
+# round-trip (~29s) on every poll. Instead, register a monitor and end the turn:
+monitor({
+  description: "CI completion for PR $PR_NUMBER",
+  command: "gh pr checks $PR_NUMBER --watch --fail-fast",
+  filter: "completed|fail|cancel"
+})
+# → end turn; the monitor's matching line arrives as an injected event.
+# For a single midpoint status peek (at most once), use:
+#   gh pr checks "$PR_NUMBER"  # one-shot, no --watch
 ```
 
 **On failure**: Get the failed run logs to understand what broke:
@@ -243,20 +251,21 @@ fi
 
 **On issues**: Cubic's review body contains structured issue descriptions. Parse them, determine which are valid (some may be false positives), and fix the valid ones per the iteration discipline below.
 
-Cubic reviews are triggered automatically on PR updates. After pushing a fix, wait for the new review to appear before checking again. Use `gh api` polling with a conditional loop:
+Cubic reviews are triggered automatically on PR updates. After pushing a fix, subscribe to the new review arriving — never spin a `for _ in $(seq 1 30)` polling loop that burns model round-trips.
 
-```bash
-# Wait for a NEW Cubic review after push. If none arrives within the bound,
-# Cubic is out of quota (or not running) → skip Gate B rather than spin forever.
+```
+# Subscribe to a NEW Cubic review after push. The monitor exits when a review
+# newer than PUSH_TIME appears, or times out (quota exhausted → Gate B SKIPPED).
 PUSH_TIME=$(date -u +%Y-%m-%dT%H:%M:%SZ)
-for _ in $(seq 1 30); do
-  LATEST_REVIEW_TIME=$(gh api "repos/${REPO}/pulls/${PR_NUMBER}/reviews" \
-    --jq '[.[] | select(.user.login == "cubic-dev-ai[bot]")] | last | .submitted_at')
-  [[ "$LATEST_REVIEW_TIME" > "$PUSH_TIME" ]] && break
-  timeout 20 gh pr checks "$PR_NUMBER" --watch >/dev/null 2>&1 || true  # spend the interval usefully
-done
-# Loop exhausted without a newer review → treat Gate B as SKIPPED (quota exhausted)
-[[ "$LATEST_REVIEW_TIME" > "$PUSH_TIME" ]] || echo "Cubic: SKIPPED (no review within bound — quota exhausted)"
+monitor({
+  description: "Cubic review for PR $PR_NUMBER",
+  command: "LATEST=$(gh api repos/${REPO}/pulls/${PR_NUMBER}/reviews --jq '[.[] | select(.user.login == "cubic-dev-ai[bot]")] | last | .submitted_at // empty'); [ -n \"$LATEST\" ] && [ \"$LATEST\" > \"$PUSH_TIME\" ] && echo NEW_REVIEW || echo WAITING",
+  filter: "NEW_REVIEW",
+  timeout_ms: 600000   # 10 min bound — if no review arrives, Gate B is SKIPPED (quota exhausted)
+})
+# → end turn; if the monitor times out without NEW_REVIEW, treat Gate B as SKIPPED.
+# For a single midpoint peek (at most once), use:
+#   gh api "repos/${REPO}/pulls/${PR_NUMBER}/reviews" --jq '[.[] | select(.user.login == "cubic-dev-ai[bot]")] | last | .submitted_at'
 ```
 
 ### Iteration discipline
@@ -292,13 +301,18 @@ gh pr merge "$PR_NUMBER" --merge --auto --delete-branch
 # are green, fall back to a direct merge: gh pr merge "$PR_NUMBER" --merge --delete-branch
 ```
 
-Then WAIT until the merge has actually completed before you report done or clean up - never walk away while the PR is still merging:
+Then subscribe to the merge completing — never block a model round-trip on an `until [ ... MERGED ]` polling loop:
 
-```bash
-# Block until the PR is actually MERGED (auto-merge lands once all required checks pass)
-until [ "$(gh pr view "$PR_NUMBER" --json state -q .state)" = "MERGED" ]; do
-  gh pr checks "$PR_NUMBER" --watch --fail-fast >/dev/null 2>&1 || true   # spend the interval on the checks
-done
+```
+# Subscribe to merge completion. The monitor exits when gh pr view returns MERGED.
+monitor({
+  description: "Merge completion for PR $PR_NUMBER",
+  command: "[ \"$(gh pr view $PR_NUMBER --json state -q .state)\" = \"MERGED\" ] && echo MERGED || echo WAITING",
+  filter: "MERGED",
+  timeout_ms: 1800000   # 30 min bound for auto-merge to land after all gates pass
+})
+# → end turn; the MERGED event wakes the session for the cleanup step.
+# If the monitor times out, check merge state once: gh pr view "$PR_NUMBER" --json state -q .state
 ```
 
 If the user opted out of merging, skip the merge but STILL run the cleanup below: the worktree is removed either way.

package/.opencode/skills/work-with-pr/SKILL.md

@@ -196,12 +196,20 @@ while true:
 
 ### Gate A: CI Checks
 
-CI is the fastest feedback loop. Wait for it to complete, then parse results.
+CI is the fastest feedback loop. Subscribe to its completion via `monitor` — never block a model round-trip on `gh pr checks --watch`.
 
-```bash
-# Wait for checks to start (GitHub needs a moment after push)
-# Then watch for completion
-gh pr checks "$PR_NUMBER" --watch --fail-fast
+```
+# Subscribe to CI completion — the monitor event wakes the session when checks finish.
+# Do NOT use `gh pr checks --watch` as a blocking tool call; it burns a full model
+# round-trip (~29s) on every poll. Instead, register a monitor and end the turn:
+monitor({
+  description: "CI completion for PR $PR_NUMBER",
+  command: "gh pr checks $PR_NUMBER --watch --fail-fast",
+  filter: "completed|fail|cancel"
+})
+# → end turn; the monitor's matching line arrives as an injected event.
+# For a single midpoint status peek (at most once), use:
+#   gh pr checks "$PR_NUMBER"  # one-shot, no --watch
 ```
 
 **On failure**: Get the failed run logs to understand what broke:
@@ -243,20 +251,21 @@ fi
 
 **On issues**: Cubic's review body contains structured issue descriptions. Parse them, determine which are valid (some may be false positives), and fix the valid ones per the iteration discipline below.
 
-Cubic reviews are triggered automatically on PR updates. After pushing a fix, wait for the new review to appear before checking again. Use `gh api` polling with a conditional loop:
+Cubic reviews are triggered automatically on PR updates. After pushing a fix, subscribe to the new review arriving — never spin a `for _ in $(seq 1 30)` polling loop that burns model round-trips.
 
-```bash
-# Wait for a NEW Cubic review after push. If none arrives within the bound,
-# Cubic is out of quota (or not running) → skip Gate B rather than spin forever.
+```
+# Subscribe to a NEW Cubic review after push. The monitor exits when a review
+# newer than PUSH_TIME appears, or times out (quota exhausted → Gate B SKIPPED).
 PUSH_TIME=$(date -u +%Y-%m-%dT%H:%M:%SZ)
-for _ in $(seq 1 30); do
-  LATEST_REVIEW_TIME=$(gh api "repos/${REPO}/pulls/${PR_NUMBER}/reviews" \
-    --jq '[.[] | select(.user.login == "cubic-dev-ai[bot]")] | last | .submitted_at')
-  [[ "$LATEST_REVIEW_TIME" > "$PUSH_TIME" ]] && break
-  timeout 20 gh pr checks "$PR_NUMBER" --watch >/dev/null 2>&1 || true  # spend the interval usefully
-done
-# Loop exhausted without a newer review → treat Gate B as SKIPPED (quota exhausted)
-[[ "$LATEST_REVIEW_TIME" > "$PUSH_TIME" ]] || echo "Cubic: SKIPPED (no review within bound — quota exhausted)"
+monitor({
+  description: "Cubic review for PR $PR_NUMBER",
+  command: "LATEST=$(gh api repos/${REPO}/pulls/${PR_NUMBER}/reviews --jq '[.[] | select(.user.login == "cubic-dev-ai[bot]")] | last | .submitted_at // empty'); [ -n \"$LATEST\" ] && [ \"$LATEST\" > \"$PUSH_TIME\" ] && echo NEW_REVIEW || echo WAITING",
+  filter: "NEW_REVIEW",
+  timeout_ms: 600000   # 10 min bound — if no review arrives, Gate B is SKIPPED (quota exhausted)
+})
+# → end turn; if the monitor times out without NEW_REVIEW, treat Gate B as SKIPPED.
+# For a single midpoint peek (at most once), use:
+#   gh api "repos/${REPO}/pulls/${PR_NUMBER}/reviews" --jq '[.[] | select(.user.login == "cubic-dev-ai[bot]")] | last | .submitted_at'
 ```
 
 ### Iteration discipline
@@ -292,13 +301,18 @@ gh pr merge "$PR_NUMBER" --merge --auto --delete-branch
 # are green, fall back to a direct merge: gh pr merge "$PR_NUMBER" --merge --delete-branch
 ```
 
-Then WAIT until the merge has actually completed before you report done or clean up - never walk away while the PR is still merging:
+Then subscribe to the merge completing — never block a model round-trip on an `until [ ... MERGED ]` polling loop:
 
-```bash
-# Block until the PR is actually MERGED (auto-merge lands once all required checks pass)
-until [ "$(gh pr view "$PR_NUMBER" --json state -q .state)" = "MERGED" ]; do
-  gh pr checks "$PR_NUMBER" --watch --fail-fast >/dev/null 2>&1 || true   # spend the interval on the checks
-done
+```
+# Subscribe to merge completion. The monitor exits when gh pr view returns MERGED.
+monitor({
+  description: "Merge completion for PR $PR_NUMBER",
+  command: "[ \"$(gh pr view $PR_NUMBER --json state -q .state)\" = \"MERGED\" ] && echo MERGED || echo WAITING",
+  filter: "MERGED",
+  timeout_ms: 1800000   # 30 min bound for auto-merge to land after all gates pass
+})
+# → end turn; the MERGED event wakes the session for the cleanup step.
+# If the monitor times out, check merge state once: gh pr view "$PR_NUMBER" --json state -q .state
 ```
 
 If the user opted out of merging, skip the merge but STILL run the cleanup below: the worktree is removed either way.

package/dist/agents/sisyphus/grok-4.d.ts

@@ -0,0 +1,20 @@
+/**
+ * Shared Grok 4.5 / Grok 4.6 Sisyphus prompt.
+ *
+ * Tuned from field experience with Grok 4.6 (Eric Zakariasson's launch field
+ * guide): phrasing intensity ("work very hard") changes nothing, and the
+ * model's own taste fills gaps well, so this is deliberately the leanest
+ * Sisyphus variant. What does move the outcome:
+ * - an explicit verification loop ("keep iterating and verifying until it's
+ *   production ready") is the single highest-leverage instruction;
+ * - a written definition of done, because otherwise the model decides done
+ *   for you;
+ * - a narration split (quiet on small changes, narrate wide-radius work) that
+ *   matches its information-dense communication style;
+ * - an anti-repetition nudge, since it duplicates component code unless asked
+ *   to break it up.
+ */
+import type { AvailableAgent, AvailableTool, AvailableSkill, AvailableCategory } from "../dynamic-agent-prompt-builder";
+import { categorizeTools } from "../dynamic-agent-prompt-builder";
+export declare function buildGrok4SisyphusPrompt(model: string, availableAgents: AvailableAgent[], availableTools?: AvailableTool[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): string;
+export { categorizeTools };

package/dist/agents/sisyphus/index.d.ts

@@ -10,6 +10,7 @@
  * - gemini.ts: Corrective overlays for Gemini's aggressive tendencies
  * - gpt-5-4.ts: Native GPT-5.4 prompt with block-structured guidance
  * - gpt-5-5.ts: Native GPT-5.5 prompt with Codex-style sections
+ * - grok-4.ts: Shared Grok 4.5/4.6 prompt, verification-loop-centered and lean
  * - kimi-k2-6.ts: Kimi K2.6 native prompt with thinking-mode stop conditions
  * - kimi-k2-7.ts: Kimi K2.7 native prompt, restrained and outcome-first
  * - kimi-k3.ts: Kimi K3 native prompt with explicit anti-overthinking calibration
@@ -22,6 +23,7 @@ export { buildClaudeFable5SisyphusPrompt } from "./claude-fable-5";
 export { buildGeminiToolMandate, buildGeminiDelegationOverride, buildGeminiVerificationOverride, buildGeminiIntentGateEnforcement, buildGeminiToolGuide, buildGeminiToolCallExamples, } from "./gemini";
 export { buildGpt54SisyphusPrompt } from "./gpt-5-4";
 export { buildGpt55SisyphusPrompt } from "./gpt-5-5";
+export { buildGrok4SisyphusPrompt } from "./grok-4";
 export { buildGlm52SisyphusPrompt } from "./glm-5-2";
 export { buildKimiK26SisyphusPrompt } from "./kimi-k2-6";
 export { buildKimiK27SisyphusPrompt } from "./kimi-k2-7";

package/dist/agents/sisyphus-agent-config.d.ts

@@ -2,4 +2,10 @@ import type { AgentConfig } from "@opencode-ai/sdk";
 import type { AgentMode } from "./types";
 export declare function buildGptSisyphusAgentConfig(mode: AgentMode, model: string, prompt: string): AgentConfig;
 export declare function buildGlmSisyphusAgentConfig(mode: AgentMode, model: string, prompt: string): AgentConfig;
+/**
+ * Grok 4.5/4.6 are xAI reasoning models: they take a reasoning effort
+ * (grok family caps allow low/medium/high) and reject Anthropic-style
+ * thinking blocks, so this is the base config plus effort only.
+ */
+export declare function buildGrokSisyphusAgentConfig(mode: AgentMode, model: string, prompt: string): AgentConfig;
 export declare function buildClaudeSisyphusAgentConfig(mode: AgentMode, model: string, prompt: string): AgentConfig;

package/dist/agents/sisyphus-agent-factory.d.ts

@@ -9,7 +9,7 @@ declare const MODE: AgentMode;
  * runtime model resolves to a different family than the configured one, the baked
  * body is the wrong family and must be rebuilt (issue #5297/#5316).
  */
-export type SisyphusPromptFamily = "kimi-k3" | "kimi-k2-7" | "kimi-k2-6" | "gpt-5-5" | "gpt-5-4" | "claude-fable-5" | "claude-opus-5" | "claude-opus-4-8" | "claude-opus-4-7" | "glm-5-2" | "fallback";
+export type SisyphusPromptFamily = "kimi-k3" | "kimi-k2-7" | "kimi-k2-6" | "gpt-5-5" | "gpt-5-4" | "claude-fable-5" | "claude-opus-5" | "claude-opus-4-8" | "claude-opus-4-7" | "glm-5-2" | "grok-4" | "fallback";
 export declare function resolveSisyphusPromptFamily(model: string): SisyphusPromptFamily;
 export declare function createSisyphusAgent(model: string, availableAgents?: AvailableAgent[], availableToolNames?: string[], availableSkills?: AvailableSkill[], availableCategories?: AvailableCategory[], useTaskSystem?: boolean): AgentConfig;
 export declare namespace createSisyphusAgent {

package/dist/agents/types.d.ts

@@ -1,6 +1,6 @@
 import type { AgentConfig } from "@opencode-ai/sdk";
-import { isClaudeFable5Model, isClaudeFableOrMythosModel, isClaudeOpus46Model, isClaudeOpus47Model, isClaudeOpus47OrLaterModel, isClaudeOpus48Model, isClaudeOpus5Model, isGeminiModel, isGlmModel, isGptModel, isKimiK2Model, isKimiK27Model, isKimiK3Model, isMiniMaxModel } from "@oh-my-opencode/model-core";
-export { isClaudeFable5Model, isClaudeFableOrMythosModel, isClaudeOpus46Model, isClaudeOpus47Model, isClaudeOpus47OrLaterModel, isClaudeOpus48Model, isClaudeOpus5Model, isGeminiModel, isGlmModel, isGptModel, isKimiK2Model, isKimiK27Model, isKimiK3Model, isMiniMaxModel, };
+import { isClaudeFable5Model, isClaudeFableOrMythosModel, isClaudeOpus46Model, isClaudeOpus47Model, isClaudeOpus47OrLaterModel, isClaudeOpus48Model, isClaudeOpus5Model, isGeminiModel, isGlmModel, isGptModel, isGrok45Model, isGrok46Model, isKimiK2Model, isKimiK27Model, isKimiK3Model, isMiniMaxModel } from "@oh-my-opencode/model-core";
+export { isClaudeFable5Model, isClaudeFableOrMythosModel, isClaudeOpus46Model, isClaudeOpus47Model, isClaudeOpus47OrLaterModel, isClaudeOpus48Model, isClaudeOpus5Model, isGeminiModel, isGlmModel, isGptModel, isGrok45Model, isGrok46Model, isKimiK2Model, isKimiK27Model, isKimiK3Model, isMiniMaxModel, };
 /**
  * Anthropic Opus 4.7+, Fable, and Mythos models reject thinking.type "enabled";
  * they require adaptive thinking plus an effort, which OpenCode core derives from