Oracle's <promise>VERIFIED</promise> arrives as a tool_result part from the
task() tool call, not as a text part. Both detectCompletionInSessionMessages
and collectAssistantText only scanned type=text parts, missing the
verification signal entirely. This caused ulw loops to fail verification
even when Oracle successfully emitted VERIFIED.
Include tool_result parts in promise detection alongside text parts.
Exclude tool_use parts to avoid false positives from instructional text.
When Oracle verification fails in ulw-loop mode, the previous Oracle
session was never aborted before restarting. Each retry created a new
descendant session, causing unbounded session accumulation and 500
errors from server overload.
Now abort the old verification session before:
- restarting the loop after failed verification
- re-entering verification phase on subsequent DONE detection
The verification_session_id was never reliably set because the
prompt-based attempt_id matching in tool-execute-after depends on
metadata.prompt surviving the delegate-task execution chain. When
this fails silently, the loop never detects Oracle's VERIFIED
emission.
Add a fallback: when exact attempt_id matching fails but oracle
agent + verification_pending state match, still set the session ID.
Add diagnostic logging to trace verification flow failures.
Add integration test covering the full verification chain.
Closes#1901
Add 'default_strategy' config option (default: 'continue') to control whether ralph-loop creates a new session per iteration ('reset') or keeps the same session ('continue'). The 'reset' strategy keeps the model in the smart zone by starting with fresh context for each iteration.
Supports --strategy flag for per-command override.
Apply the internal initiator marker to automated continuation, recovery, babysitter, stop-hook, and hook-message injections so Copilot attribution consistently sets x-initiator=agent for system-generated prompts.
Reasoning parts could contain completion-like text triggering false
positives. Also handles session.messages returning either an array
or {data: [...]} shape.
The ralph-loop hook calls promptAsync in the implementation, but the
test mock only defined prompt(). Added promptAsync with identical
behavior to make tests pass.
- All 38 ralph-loop tests now pass
- Total test suite: 2361 pass, 3 fail (unrelated to this change)
- Add timing.ts module for test-only timing configuration
- Replace hardcoded wait times with getTimingConfig()
- Enable all previously skipped tests (ralph-loop, session-state, delegate-task)
- Tests now complete in ~2s instead of timing out
* fix(ralph-loop): skip user messages in transcript completion detection (#622)
The transcript-based completion detection was searching the entire JSONL
file for <promise>DONE</promise>, including user message entries. The
RALPH_LOOP_TEMPLATE instructional text contains this literal pattern,
which gets recorded as a user message, causing false positive completion
detection on every iteration. This made the loop always terminate at
iteration 1.
Fix: Parse JSONL entries line-by-line and skip entries with type 'user'
so only tool_result/assistant entries are checked for the completion
promise. Also remove the hardcoded <promise>DONE</promise> from the
template exit conditions as defense-in-depth.
* chore: changes by sisyphus-dev-ai
---------
Co-authored-by: sisyphus-dev-ai <sisyphus-dev-ai@users.noreply.github.com>
OpenCode API returns different structures for user vs assistant messages:
- User: info.model = { providerID, modelID } (nested)
- Assistant: info.modelID, info.providerID (flat top-level)
Previous code only checked nested format, causing model info loss when
continuation hooks fired after assistant messages.
Files modified:
- todo-continuation-enforcer.ts
- ralph-loop/index.ts
- sisyphus-task/tools.ts
- background-agent/manager.ts
Added test for assistant message model extraction.
- Add shared/system-directive.ts with SYSTEM_DIRECTIVE_PREFIX constant
- Unify all system message prefixes to [SYSTEM DIRECTIVE: OH-MY-OPENCODE - ...]
- Add isSystemDirective() filter to keyword-detector to skip system messages
- Update prometheus-md-only tests to use new prefix constants
Previously, continuation hooks (todo-continuation, boulder-continuation, ralph-loop)
and background tasks resolved model info from filesystem cache, which could be stale
or missing. This caused session.prompt to fallback to default model (Sonnet) instead
of using the originally configured model (e.g., Opus).
Now all session.prompt calls first try API (session.messages) to get current model
info, with filesystem as fallback if API fails.
Affected files:
- todo-continuation-enforcer.ts
- sisyphus-orchestrator/index.ts
- ralph-loop/index.ts
- background-agent/manager.ts
- sisyphus-task/tools.ts
- hook-message-injector/index.ts (export ToolPermission type)
Instead of using stale parentModel/parentAgent from task state, now
dynamically looks up the current message to get fresh model/agent values.
Applied across all prompt injection points:
- background-agent notifyParentSession
- ralph-loop continuation
- sisyphus-orchestrator boulder continuation
- sisyphus-task resume
🤖 Generated with [OhMyOpenCode](https://github.com/code-yeongyu/oh-my-opencode) assistance
- Add test for starting new loop while previous loop active (different session)
- Add test for restarting loop in same session
- Verifies startLoop properly overwrites state and resets iteration
* fix(ralph-loop): clear orphaned state when original session no longer exists
When a session with an active ralph-loop terminates abnormally (abort, window close),
the state file remains with active: true. Previously, when a new session started,
the hook would skip the orphaned state without cleaning it up.
This fix adds session existence validation:
- Before skipping a loop owned by a different session, check if that session still exists
- If the original session no longer exists, clear the orphan state and log
- If the original session still exists, skip as before (it's another active session's loop)
Changes:
- Add checkSessionExists option to RalphLoopOptions for dependency injection
- Wire up sessionExists from session-manager as the default implementation
- Add tests for orphan state cleanup and active session preservation
* fix(ralph-loop): add error handling around checkSessionExists call
Wraps the async checkSessionExists call in try/catch for consistency
with other async operations in this file. If the check throws, logs
the error and falls back to the original behavior (not clearing state).
---------
Co-authored-by: sisyphus-dev-ai <sisyphus-dev-ai@users.noreply.github.com>
* fix(ralph-loop): detect completion promise from session messages API
The completion promise (e.g., <promise>DONE</promise>) was not being detected
because assistant text messages were never recorded to the transcript file.
Only user messages, tool uses, and tool results were recorded.
This fix adds a new detection method that fetches session messages via the
OpenCode API and checks assistant text messages for the completion promise.
The transcript file check is kept as a fallback.
Fixes#412
* refactor(ralph-loop): address review feedback
- Simplify response parsing to use response.data consistently (greptile)
- Add session ID tracking to messages mock for better test coverage (cubic)
- Add assertion to verify correct session ID is passed to API
---------
Co-authored-by: sisyphus-dev-ai <sisyphus-dev-ai@users.noreply.github.com>
YeonGyu-Kim