74 lines
3.2 KiB
Markdown
74 lines
3.2 KiB
Markdown
---
|
|
description: Maintain semantic integrity via multi-agent delegation. Analyzes codebase, delegates markup tasks to Semantic Engineer, verifies via Reviewer Agent, and reports status.
|
|
---
|
|
|
|
## User Input
|
|
|
|
```text
|
|
$ARGUMENTS
|
|
```
|
|
|
|
You **MUST** consider the user input before proceeding (if not empty).
|
|
|
|
## Goal
|
|
|
|
Ensure the codebase 100% adheres to the semantic standards defined in `.ai/standards/semantics.md` (GRACE-Poly Protocol). You are the **Manager/Supervisor**. You do not write code. You manage the queue, delegate files to the Semantic Engineer, audit their work via the Reviewer Agent, and commit successful changes.
|
|
|
|
## Operating Constraints
|
|
|
|
1. **ROLE: Orchestrator**: High-level coordination ONLY. Do not output raw code diffs yourself.
|
|
2. **DELEGATION PATTERN**: Strict `Orchestrator -> Engineer -> Reviewer -> Orchestrator` loop.
|
|
3. **FAIL-FAST METRICS**: If the Reviewer Agent rejects a file 3 times in a row, drop the file from the current queue and mark it as `[HUMAN_INTERVENTION_REQUIRED]`.
|
|
4. **TIER AWARENESS**: CRITICAL files MUST be processed first. A failure in a CRITICAL file blocks the entire pipeline.
|
|
|
|
## Execution Steps
|
|
|
|
### 1. Generate Semantic State (Analyze)
|
|
Run the generator script to map the current reality:
|
|
```bash
|
|
python3 generate_semantic_map.py
|
|
```
|
|
Parse the output (Global Score, Critical Parsing Errors, Files with Score < 0.7).
|
|
|
|
### 2. Formulate Task Queue
|
|
Create an execution queue based on the report. Priority:
|
|
- **Priority 1 (Blockers)**: Files with "Critical Parsing Errors" (unclosed `[/DEF]` anchors).
|
|
- **Priority 2 (Tier 1)**: `CRITICAL` tier modules missing mandatory tags (`@PRE`, `@POST`, `belief_scope`).
|
|
- **Priority 3 (Tier 2)**: `STANDARD` modules with missing graph relations (`@RELATION`).
|
|
|
|
### 3. The Delegation Loop (For each file in the queue)
|
|
For every target file, execute this exact sequence:
|
|
|
|
* **Step 3A (Delegate to Worker):** Send the file path and the specific violation from the report to the **Semantic Markup Agent (Engineer)**.
|
|
*Prompt*: `"Fix semantic violations in [FILE]. Current issues: [ISSUES]. Apply GRACE-Poly standards without changing business logic."*
|
|
* **Step 3B (Delegate to Auditor):** Once the Engineer returns the modified file, send it to the **Reviewer Agent (Auditor)**.
|
|
*Prompt*: `"Verify GRACE-Poly compliance for [FILE]. Check for paired [DEF] anchors, complete contracts, and belief_scope usage. Return PASS or FAIL with specific line errors."*
|
|
* **Step 3C (Evaluate):**
|
|
* If Auditor returns `PASS`: Apply the diff to the codebase. Move to the next file.
|
|
* If Auditor returns `FAIL`: Send the Auditor's error report back to the Engineer (Step 3A). Repeat max 3 times.
|
|
|
|
### 4. Verification
|
|
Once the queue is empty, re-run `python3 generate_semantic_map.py` to prove the metrics have improved.
|
|
|
|
## Output Format
|
|
|
|
Return a structured summary of the operation:
|
|
|
|
```text
|
|
=== GRACE SEMANTIC ORCHESTRATION REPORT ===
|
|
Initial Global Score: [X]%
|
|
Final Global Score: [Y]%
|
|
Status:[PASS / BLOCKED]
|
|
|
|
Files Processed:
|
|
1. [file_path] -[PASS (1 attempt) | PASS (2 attempts) | FAILED]
|
|
2. ...
|
|
|
|
Escalations (Human Intervention Required):
|
|
-[file_path]: Failed auditor review 3 times. Reason: [Last Auditor Note].
|
|
```
|
|
## Context
|
|
|
|
$ARGUMENTS
|
|
```
|