45dfc4ec66
- VERIFICATION_REMINDER: add Step 2 manual code review (non-negotiable) - Require Read of EVERY changed file line by line - Cross-check subagent claims vs actual code - Verify logic correctness, completeness, edge cases, patterns - Add Step 5: direct boulder state check via Read plan file - Count remaining tasks directly, no cached state - BOULDER_CONTINUATION_PROMPT: add first rule to read plan file immediately - verification-reminders.ts: restructure steps 5-8 for boulder/todo checks - Atlas default.ts (Claude): enhance 3.4 QA with A/B/C/D sections - A: Automated verification - B: Manual code review (non-negotiable) - C: Hands-on QA (if applicable) - D: Check boulder state directly - Atlas gpt.ts (GPT-5.2): apply same QA enhancements with GPT-optimized structure - verification_rules: update both Claude and GPT versions with manual review requirements Addresses issue where Atlas would skip manual code inspection after delegation, leading to rubber-stamping of broken or incomplete work.
178 lines
6.5 KiB
TypeScript
178 lines
6.5 KiB
TypeScript
import { createSystemDirective, SystemDirectiveTypes } from "../../shared/system-directive"
|
|
|
|
export const DIRECT_WORK_REMINDER = `
|
|
|
|
---
|
|
|
|
${createSystemDirective(SystemDirectiveTypes.DELEGATION_REQUIRED)}
|
|
|
|
You just performed direct file modifications outside \`.sisyphus/\`.
|
|
|
|
**You are an ORCHESTRATOR, not an IMPLEMENTER.**
|
|
|
|
As an orchestrator, you should:
|
|
- **DELEGATE** implementation work to subagents via \`task\`
|
|
- **VERIFY** the work done by subagents
|
|
- **COORDINATE** multiple tasks and ensure completion
|
|
|
|
You should NOT:
|
|
- Write code directly (except for \`.sisyphus/\` files like plans and notepads)
|
|
- Make direct file edits outside \`.sisyphus/\`
|
|
- Implement features yourself
|
|
|
|
**If you need to make changes:**
|
|
1. Use \`task\` to delegate to an appropriate subagent
|
|
2. Provide clear instructions in the prompt
|
|
3. Verify the subagent's work after completion
|
|
|
|
---
|
|
`
|
|
|
|
export const BOULDER_CONTINUATION_PROMPT = `${createSystemDirective(SystemDirectiveTypes.BOULDER_CONTINUATION)}
|
|
|
|
You have an active work plan with incomplete tasks. Continue working.
|
|
|
|
RULES:
|
|
- **FIRST**: Read the plan file NOW to check exact current progress — count remaining \`- [ ]\` tasks
|
|
- Proceed without asking for permission
|
|
- Change \`- [ ]\` to \`- [x]\` in the plan file when done
|
|
- Use the notepad at .sisyphus/notepads/{PLAN_NAME}/ to record learnings
|
|
- Do not stop until all tasks are complete
|
|
- If blocked, document the blocker and move to the next task`
|
|
|
|
export const VERIFICATION_REMINDER = `**MANDATORY: WHAT YOU MUST DO RIGHT NOW**
|
|
|
|
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
|
|
|
|
CRITICAL: Subagents FREQUENTLY LIE about completion.
|
|
Tests FAILING, code has ERRORS, implementation INCOMPLETE - but they say "done".
|
|
|
|
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
|
|
|
|
**STEP 1: AUTOMATED VERIFICATION (DO THIS FIRST)**
|
|
|
|
Run these commands YOURSELF - do NOT trust agent's claims:
|
|
1. \`lsp_diagnostics\` on changed files → Must be CLEAN
|
|
2. \`bash\` to run tests → Must PASS
|
|
3. \`bash\` to run build/typecheck → Must succeed
|
|
|
|
**STEP 2: MANUAL CODE REVIEW (NON-NEGOTIABLE — DO NOT SKIP)**
|
|
|
|
Automated checks are NECESSARY but INSUFFICIENT. You MUST read the actual code.
|
|
|
|
**RIGHT NOW — \`Read\` EVERY file the subagent touched. No exceptions.**
|
|
|
|
For EACH changed file, verify:
|
|
1. Does the implementation logic ACTUALLY match the task requirements?
|
|
2. Are there incomplete stubs (TODO comments, placeholder code, hardcoded values)?
|
|
3. Are there logic errors, off-by-one bugs, or missing edge cases?
|
|
4. Does it follow existing codebase patterns and conventions?
|
|
5. Are imports correct? No unused or missing imports?
|
|
6. Is error handling present where needed?
|
|
|
|
**Cross-check the subagent's claims against reality:**
|
|
- Subagent said "Updated X" → READ X. Is it actually updated?
|
|
- Subagent said "Added tests" → READ tests. Do they test the RIGHT behavior?
|
|
- Subagent said "Follows patterns" → COMPARE with reference. Does it actually?
|
|
|
|
**If you cannot explain what the changed code does, you have not reviewed it.**
|
|
**If you skip this step, you are rubber-stamping broken work.**
|
|
|
|
**STEP 3: DETERMINE IF HANDS-ON QA IS NEEDED**
|
|
|
|
| Deliverable Type | QA Method | Tool |
|
|
|------------------|-----------|------|
|
|
| **Frontend/UI** | Browser interaction | \`/playwright\` skill |
|
|
| **TUI/CLI** | Run interactively | \`interactive_bash\` (tmux) |
|
|
| **API/Backend** | Send real requests | \`bash\` with curl |
|
|
|
|
Static analysis CANNOT catch: visual bugs, animation issues, user flow breakages.
|
|
|
|
**STEP 4: IF QA IS NEEDED - ADD TO TODO IMMEDIATELY**
|
|
|
|
\`\`\`
|
|
todowrite([
|
|
{ id: "qa-X", content: "HANDS-ON QA: [specific verification action]", status: "pending", priority: "high" }
|
|
])
|
|
\`\`\`
|
|
|
|
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
|
|
|
|
**BLOCKING: DO NOT proceed until Steps 1-4 are ALL completed.**
|
|
**Skipping Step 2 (manual code review) = unverified work = FAILURE.**`
|
|
|
|
export const ORCHESTRATOR_DELEGATION_REQUIRED = `
|
|
|
|
---
|
|
|
|
${createSystemDirective(SystemDirectiveTypes.DELEGATION_REQUIRED)}
|
|
|
|
**STOP. YOU ARE VIOLATING ORCHESTRATOR PROTOCOL.**
|
|
|
|
You (Atlas) are attempting to directly modify a file outside \`.sisyphus/\`.
|
|
|
|
**Path attempted:** $FILE_PATH
|
|
|
|
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
|
|
|
|
**THIS IS FORBIDDEN** (except for VERIFICATION purposes)
|
|
|
|
As an ORCHESTRATOR, you MUST:
|
|
1. **DELEGATE** all implementation work via \`task\`
|
|
2. **VERIFY** the work done by subagents (reading files is OK)
|
|
3. **COORDINATE** - you orchestrate, you don't implement
|
|
|
|
**ALLOWED direct file operations:**
|
|
- Files inside \`.sisyphus/\` (plans, notepads, drafts)
|
|
- Reading files for verification
|
|
- Running diagnostics/tests
|
|
|
|
**FORBIDDEN direct file operations:**
|
|
- Writing/editing source code
|
|
- Creating new files outside \`.sisyphus/\`
|
|
- Any implementation work
|
|
|
|
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
|
|
|
|
**IF THIS IS FOR VERIFICATION:**
|
|
Proceed if you are verifying subagent work by making a small fix.
|
|
But for any substantial changes, USE \`task\`.
|
|
|
|
**CORRECT APPROACH:**
|
|
\`\`\`
|
|
task(
|
|
category="...",
|
|
prompt="[specific single task with clear acceptance criteria]"
|
|
)
|
|
\`\`\`
|
|
|
|
DELEGATE. DON'T IMPLEMENT.
|
|
|
|
---
|
|
`
|
|
|
|
export const SINGLE_TASK_DIRECTIVE = `
|
|
|
|
${createSystemDirective(SystemDirectiveTypes.SINGLE_TASK_ONLY)}
|
|
|
|
**STOP. READ THIS BEFORE PROCEEDING.**
|
|
|
|
If you were NOT given **exactly ONE atomic task**, you MUST:
|
|
1. **IMMEDIATELY REFUSE** this request
|
|
2. **DEMAND** the orchestrator provide a single, specific task
|
|
|
|
**Your response if multiple tasks detected:**
|
|
> "I refuse to proceed. You provided multiple tasks. An orchestrator's impatience destroys work quality.
|
|
>
|
|
> PROVIDE EXACTLY ONE TASK. One file. One change. One verification.
|
|
>
|
|
> Your rushing will cause: incomplete work, missed edge cases, broken tests, wasted context."
|
|
|
|
**WARNING TO ORCHESTRATOR:**
|
|
- Your hasty batching RUINS deliverables
|
|
- Each task needs FULL attention and PROPER verification
|
|
- Batch delegation = sloppy work = rework = wasted tokens
|
|
|
|
**REFUSE multi-task requests. DEMAND single-task clarity.**
|
|
`
|