fix(atlas): schedule delayed retry when cooldown blocks boulder continuation by code-yeongyu · Pull Request #2335 · code-yeongyu/oh-my-openagent

code-yeongyu · 2026-03-06T07:14:44Z

Summary

Fixes boulder continuation getting permanently stuck when the model's response is immediately aborted after atlas injects a continuation prompt
Root cause: After abort, OpenCode fires a burst of session.idle events within milliseconds, all blocked by 5s cooldown. After the burst stops, no more idle events come (state-change based, not periodic) — session stuck forever
Fix: Schedule a one-shot setTimeout (cooldown + 1s = 6s) when cooldown blocks an idle event for a boulder with incomplete plan. Timer re-checks boulder state/plan/continuation-stopped before injecting. Deduped via pendingRetryTimer field. Cleaned up on session.deleted/session.compacted

Changes

src/hooks/atlas/types.ts: Added pendingRetryTimer field to SessionState
src/hooks/atlas/event-handler.ts: Added delayed retry timer in cooldown skip branch + timer cleanup in session.deleted/compacted handlers
src/hooks/atlas/index.test.ts: 5 new TDD tests covering retry scheduling, dedup, plan-complete bailout, and cleanup on delete/compact

Test Results

All 46 tests pass (41 existing + 5 new). Typecheck clean.

Summary by cubic

Prevents atlas boulder continuation from getting stuck after an aborted response by scheduling a delayed retry once the cooldown passes. Also tightens retry checks, adds model‑routed prompts, the Final Verification Wave, QA scenario checks, and updates fallbacks to prefer gpt‑5.4.

Bug Fixes
- Atlas: schedule a 6s delayed retry when cooldown blocks idle, preventing permanent stalls after abort-driven idle bursts.
- Retry callback: re-checks failure backoff count, boulder session membership, and running tasks before injecting; one timer per session; cleared on session.deleted/compacted.
- Tests: use fake timers to intercept delayed retries and speed up runs; added coverage for retry scheduling, dedup, bailout on completed plans, and timer cleanup.
- Start‑Work: remove “Worktree Setup Required” when unspecified; when a worktree exists, inject a concise “Worktree Active” block with strict path rules.
New Features
- Hephaestus agent: new GPT‑specific prompt variants (gpt‑5.4, gpt‑5.3‑codex, generic) with auto routing and tests.
- Sisyphus‑Junior: adds GPT‑5.4 and GPT‑5.3‑codex prompts with model‑based routing.
- Atlas prompts: add Final Verification Wave tracking and todos.
- Metis & Momus: enforce executable QA scenarios for each task.
- Model selection: “unspecified‑high” now prefers gpt‑5.4; fallbacks updated and add opencode provider to glm‑5; CLI snapshots aligned.

^{Written for commit b062fc4. Summary will update on new commits.}

…trong worktree active instructions When no worktree is specified in boulder, stop injecting 'Worktree Setup Required' instructions. When worktree IS present, inject emphatic instructions ensuring the agent and all subagents operate exclusively within the worktree directory.

…-5/k2p5/kimi-k2.5 fallbacks 🤖 Generated with assistance of [OhMyOpenCode](https://github.com/code-yeongyu/oh-my-opencode) Co-authored-by: Sisyphus <[email protected]>

🤖 Generated with assistance of [OhMyOpenCode](https://github.com/code-yeongyu/oh-my-opencode) Co-authored-by: Sisyphus <[email protected]>

…, gpt-5-3-codex, generic gpt) Split GPT prompt into three variants with model-based routing: - gpt-5-4.ts: GPT-5.4 optimized (expert coding agent framing, prose-first) - gpt-5-3-codex.ts: GPT-5.3-Codex optimized (Hephaestus-style Senior Engineer) - gpt.ts: Generic GPT fallback (Hephaestus-style, for any other GPT model) Routing: gpt-5.4 → gpt-5-4 | gpt-5.3-codex → gpt-5-3-codex | other GPT → gpt 🤖 Generated with assistance of [OhMyOpenCode](https://github.com/code-yeongyu/oh-my-opencode)

… routing Split monolithic hephaestus.ts into directory with model-specific prompt variants (gpt-5-4.ts, gpt-5-3-codex.ts, gpt.ts) mirroring the sisyphus-junior pattern. Generic gpt.ts uses pre-codex-tuning prompt as fallback for non-specific GPT models. Also adds isGpt5_4Model and isGpt5_3CodexModel helpers to types.ts.

…nuation When atlas injects a boulder continuation via promptAsync() and the model's response is immediately aborted (MessageAbortedError), OpenCode fires a burst of session.idle events within milliseconds. Atlas blocks all of them due to the 5-second cooldown. After the burst, OpenCode stops generating session.idle events (it's state-change based, not periodic), leaving the session stuck forever. Fix: When cooldown blocks an idle event for a boulder session with an incomplete plan, schedule a one-shot setTimeout (cooldown + 1s) to re-attempt injection. The timer callback re-checks boulder state, plan progress, and continuation-stopped flag before injecting. Only one timer per session is allowed (deduped via pendingRetryTimer field). Timers are cleaned up on session.deleted and session.compacted events.

cubic-dev-ai

4 issues found across 40 files

Confidence score: 3/5

There is a concrete regression risk in src/hooks/atlas/event-handler.ts: delayed retry can continue without re-running full eligibility checks, which may cause duplicate or invalid injections in real flows.
src/cli/model-fallback-requirements.ts is missing the opencode provider for glm-5, creating a runtime/requirements mismatch risk that can surface as provider fallback failures.
This lands at moderate merge risk because multiple medium-to-high severity issues are identified with high confidence, including one clearly user-impacting behavior and several maintainability/CI stability concerns.
Pay close attention to src/hooks/atlas/event-handler.ts, src/cli/model-fallback-requirements.ts, src/hooks/atlas/index.test.ts, and src/agents/sisyphus-junior/gpt-5-3-codex.ts - injection validity, provider consistency, test-runtime slowdown, and duplicated template logic need follow-up.

Prompt for AI agents (unresolved issues)


Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="src/cli/model-fallback-requirements.ts">

<violation number="1" location="src/cli/model-fallback-requirements.ts:275">
P2: Add the 'opencode' provider to the glm-5 model to remain consistent with visual-engineering and runtime requirements.</violation>
</file>

<file name="src/hooks/atlas/index.test.ts">

<violation number="1" location="src/hooks/atlas/index.test.ts:1449">
P2: Using real `setTimeout` with a 7000ms delay in unit tests adds 35 seconds of dead waiting time across these 5 new tests, significantly slowing down CI and local test execution. Consider mocking the timer (e.g., using `jest.useFakeTimers()` provided by `bun:test`, or a custom fake timer like the one used in `todo-continuation-enforcer.test.ts`) to instantly advance time instead.</violation>
</file>

<file name="src/agents/sisyphus-junior/gpt-5-3-codex.ts">

<violation number="1" location="src/agents/sisyphus-junior/gpt-5-3-codex.ts:11">
P2: This file is an exact duplicate of \`src/agents/sisyphus-junior/gpt.ts\`, with only the function names and top comment changed. The entire 150-line string template is duplicated verbatim. Consider reusing \`buildGptSisyphusJuniorPrompt\` or extracting the shared prompt string to avoid maintaining two identical copies.</violation>
</file>

<file name="src/hooks/atlas/event-handler.ts">

<violation number="1" location="src/hooks/atlas/event-handler.ts:128">
P1: Delayed retry injects continuation without re-running full eligibility checks, which can trigger duplicate or invalid injections.</violation>
</file>

_{Reply with feedback, questions, or to request a fix. Tag @cubic-dev-ai to re-run a review.}

cubic-dev-ai · 2026-03-06T07:23:21Z

@@ -0,0 +1,151 @@
+/**


P2: This file is an exact duplicate of `src/agents/sisyphus-junior/gpt.ts`, with only the function names and top comment changed. The entire 150-line string template is duplicated verbatim. Consider reusing `buildGptSisyphusJuniorPrompt` or extracting the shared prompt string to avoid maintaining two identical copies.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At src/agents/sisyphus-junior/gpt-5-3-codex.ts, line 11: <comment>This file is an exact duplicate of \`src/agents/sisyphus-junior/gpt.ts\`, with only the function names and top comment changed. The entire 150-line string template is duplicated verbatim. Consider reusing \`buildGptSisyphusJuniorPrompt\` or extracting the shared prompt string to avoid maintaining two identical copies.</comment> <file context> @@ -0,0 +1,151 @@ + +import { resolvePromptAppend } from "../builtin-agents/resolve-file-uri" + +export function buildGpt53CodexSisyphusJuniorPrompt( + useTaskSystem: boolean, + promptAppend?: string </file context>

Address Cubic P1 review: timer callback now re-checks failure backoff count, boulder session membership, and running background tasks before injecting continuation, matching the main idle handler's eligibility gate.

…vider to glm-5 Replace real setTimeout(7000) with fake timer interception in atlas retry tests (35s -> 227ms). Add missing opencode provider to glm-5 fallback in unspecified-high category.

code-yeongyu and others added 8 commits March 6, 2026 14:20

refactor(atlas): add Final Verification Wave to orchestration workflow

cde6566

🤖 Generated with assistance of [OhMyOpenCode](https://github.com/code-yeongyu/oh-my-opencode) Co-authored-by: Sisyphus <[email protected]>

feat(metis,momus): add QA scenario executability checks

17f11a5

🤖 Generated with assistance of [OhMyOpenCode](https://github.com/code-yeongyu/oh-my-opencode) Co-authored-by: Sisyphus <[email protected]>

chore: rebuild platform binaries

533aa6d

🤖 Generated with assistance of [OhMyOpenCode](https://github.com/code-yeongyu/oh-my-opencode) Co-authored-by: Sisyphus <[email protected]>

cubic-dev-ai bot reviewed Mar 6, 2026

View reviewed changes

code-yeongyu added 2 commits March 6, 2026 16:31

fix(atlas): add full eligibility checks to delayed retry callback

4eb38d9

Address Cubic P1 review: timer callback now re-checks failure backoff count, boulder session membership, and running background tasks before injecting continuation, matching the main idle handler's eligibility gate.

fix: address Cubic P2 review - fake timers in tests, add opencode pro…

b062fc4

…vider to glm-5 Replace real setTimeout(7000) with fake timer interception in atlas retry tests (35s -> 227ms). Add missing opencode provider to glm-5 fallback in unspecified-high category.

code-yeongyu merged commit cfb9435 into dev Mar 6, 2026
8 checks passed

code-yeongyu deleted the fix/boulder-continuation-abort branch March 6, 2026 08:00

germamsatere-sudo approved these changes Mar 7, 2026

View reviewed changes

HaD0Yun mentioned this pull request Mar 17, 2026

feat(atlas): persist preferred task session reuse #2640

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(atlas): schedule delayed retry when cooldown blocks boulder continuation#2335

fix(atlas): schedule delayed retry when cooldown blocks boulder continuation#2335
code-yeongyu merged 10 commits intodevfrom
fix/boulder-continuation-abort

code-yeongyu commented Mar 6, 2026 •

edited by cubic-dev-ai bot

Loading

Uh oh!

cubic-dev-ai bot left a comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cubic-dev-ai bot Mar 6, 2026 •

edited

Loading

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

code-yeongyu commented Mar 6, 2026 • edited by cubic-dev-ai bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Changes

Test Results

Summary by cubic

Uh oh!

cubic-dev-ai bot left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cubic-dev-ai bot Mar 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

code-yeongyu commented Mar 6, 2026 •

edited by cubic-dev-ai bot

Loading

cubic-dev-ai bot Mar 6, 2026 •

edited

Loading