commit 151075d33c16a3591dd6c57a235e0a3aa9ccc787
Author: Ben Sima <ben@bensima.com>
Date: Mon Dec 1 07:58:14 2025
Improve Jr agent structure with progress file and incremental workflow
Perfect! All changes are in place and working correctly. Let me
create a
I have successfully implemented the improvements to Jr Worker
agent stru
1. **Progress File Tracking**
- Added `readProgressFile` function to read
`_/llm/${taskId}-progress - Added `buildProgressPrompt` function
to include progress context in - Modified `runWithEngine` to load
and include progress at the start
2. **Incremental Workflow Enforcement**
- Updated base prompt to explicitly instruct: "Pick ONE specific
chan - Added "INCREMENTAL WORKFLOW (IMPORTANT)" section with clear
guidanc - Added instruction to write progress after each change -
Emphasized that tasks may be run multiple times to complete all cha
3. **Explicit Verification**
- Maintained existing requirement to run `bild --test` before
complet - Added instruction to save progress only after tests pass -
Clarified that code must be left in clean, testable state
4. **Avoid Redundant Testing**
- Updated BUILD SYSTEM NOTES to clarify running `bild --test`
on name - Added explicit instruction not to re-run tests unless
more changes - Explained that bild handles dependencies transitively
- `bild --test Omni/Agent/Worker.hs` - **PASSED** ✓ - `lint
Omni/Agent/Worker.hs` - **NO ISSUES** ✓
- `_/llm/t-203-progress.md` - Progress file documenting this
implementat - `_/llm/t-203-implementation-summary.md` - Detailed
summary of changes
The implementation follows industry best practices from Anthropic,
OpenA - Reduced token usage through focused, incremental changes
- Better code quality with isolated, tested changes - Improved
reliability with progress tracking across sessions - Clear workflow
preventing "declaring victory" too early
Task-Id: t-203
diff --git a/Omni/Agent/Worker.hs b/Omni/Agent/Worker.hs
index cc5f7300..38e29cb4 100644
--- a/Omni/Agent/Worker.hs
+++ b/Omni/Agent/Worker.hs
@@ -209,14 +209,20 @@ runWithEngine worker repo task = do
-- Check for retry context
maybeRetry <- TaskCore.getRetryContext (TaskCore.taskId task)
+ -- Read progress file if it exists
+ progressContent <- readProgressFile repo (TaskCore.taskId task)
+
-- Build the full prompt
let ns = fromMaybe "." (TaskCore.taskNamespace task)
let basePrompt = buildBasePrompt task ns repo
+ -- Add progress context if present
+ let progressPrompt = buildProgressPrompt progressContent
+
-- Add retry context if present
let retryPrompt = buildRetryPrompt maybeRetry
- let prompt = basePrompt <> retryPrompt
+ let prompt = basePrompt <> progressPrompt <> retryPrompt
-- Read AGENTS.md
agentsMd <-
@@ -314,22 +320,33 @@ buildBasePrompt task ns repo =
<> "Your goal is to implement the following task:\n\n"
<> formatTask task
<> "\n\nCRITICAL INSTRUCTIONS:\n"
- <> "1. Analyze the codebase to understand where to make changes.\n"
- <> "2. Implement the changes by editing files.\n"
- <> "3. BEFORE finishing, you MUST run: bild --test "
+ <> "1. Read AGENTS.md and any existing progress file for this task.\n"
+ <> "2. Pick ONE specific change to implement (not everything at once).\n"
+ <> "3. Analyze the codebase to understand where to make that change.\n"
+ <> "4. Implement ONLY that one change.\n"
+ <> "5. BEFORE finishing, you MUST run: bild --test "
<> ns
<> "\n"
- <> "4. Fix ALL errors from bild --test (including lint issues).\n"
- <> "5. Keep running bild --test until it passes with no errors.\n"
- <> "6. Do NOT update task status or manage git.\n"
- <> "7. Only exit after bild --test passes.\n\n"
+ <> "6. Fix ALL errors from bild --test (including lint issues).\n"
+ <> "7. Keep running bild --test until it passes with no errors.\n"
+ <> "8. After tests pass, write progress to: _/llm/"
+ <> TaskCore.taskId task
+ <> "-progress.md\n"
+ <> "9. Do NOT update task status or manage git.\n"
+ <> "10. Only exit after bild --test passes and progress is saved.\n\n"
+ <> "INCREMENTAL WORKFLOW (IMPORTANT):\n"
+ <> "- DO NOT try to implement everything in one go\n"
+ <> "- Make ONE focused change, test it, save progress, then stop\n"
+ <> "- The task may be run multiple times to complete all changes\n"
+ <> "- Each session should leave the code in a clean, testable state\n"
+ <> "- If the task is already complete, just verify tests pass and note that in progress\n\n"
<> "IMPORTANT: The git commit will fail if lint finds issues.\n"
<> "You must fix all lint suggestions.\n\n"
<> "BUILD SYSTEM NOTES:\n"
<> "- Running 'bild --test "
<> ns
<> "' automatically tests ALL dependencies of that namespace\n"
- <> "- You do NOT need to run bild --test on individual files - just the main namespace\n"
+ <> "- You do NOT need to run bild --test on individual files - just the main namespace ONCE\n"
<> "- Once tests pass, do NOT re-run them unless you make more changes\n"
<> "- The 'lint' command will be run automatically during git commit via hooks\n"
<> "- You can run 'lint --fix' on changed files if needed, but it's optional\n\n"
@@ -341,6 +358,29 @@ buildBasePrompt task ns repo =
<> ns
<> "\n"
+-- | Read progress file for a task if it exists
+readProgressFile :: FilePath -> Text -> IO (Maybe Text)
+readProgressFile repo taskId = do
+ let progressPath = repo </> "_" </> "llm" </> Text.unpack taskId <> "-progress.md"
+ exists <- Directory.doesFileExist progressPath
+ if exists
+ then Just </ readFile progressPath
+ else pure Nothing
+
+-- | Build progress context prompt
+buildProgressPrompt :: Maybe Text -> Text
+buildProgressPrompt Nothing = ""
+buildProgressPrompt (Just progress) =
+ "\n\n## PROGRESS FROM PREVIOUS SESSIONS\n\n"
+ <> "This task has been worked on before. Here's what has been completed:\n\n"
+ <> progress
+ <> "\n\n"
+ <> "IMPORTANT:\n"
+ <> "- Review this progress to understand what's already done\n"
+ <> "- Do NOT repeat work that's already completed\n"
+ <> "- Pick the NEXT logical step that hasn't been done yet\n"
+ <> "- Update the progress file after completing your change\n\n"
+
-- | Build retry context prompt
buildRetryPrompt :: Maybe TaskCore.RetryContext -> Text
buildRetryPrompt Nothing = ""